WO2019036749A1

WO2019036749A1 - A system for detecting early parkinson's disease and other neurological diseases and movement disorders

Info

Publication number: WO2019036749A1
Application number: PCT/AU2018/050590
Authority: WO
Inventors: Warwick Russell ADAMS
Original assignee: Adams Warwick Russell
Priority date: 2017-08-24
Filing date: 2018-06-13
Publication date: 2019-02-28

Abstract

Disclosed is a novel system (100) and method for determining the presence of movement related disorders such as Parkinson's disease through the use of patient interaction with a data input device (104), from which characteristics indicative of movement related disorders may be determined. The system (100) comprises an input device (104) such as a keyboard, through which a patient can input data through the patients hand (112). The input device (104) is operably connected to a processor (106), memory (108) and output (110). The output (110) producing a result based on the patients keystroke inputs through the patients hand (112). The system uses machine learning, statistical analysis and regression analysis to determine among other things bradykinesia, tremor and Parkinson's disease.

Description

A SYS TE M F OR D ET E CTING E AR LY PAR KINS ON'S DIS EAS E AND OT H E R NE U R OL OGICAL DIS EAS E S AND MOVE ME NT DIS OR DE RS

F ield

[1] The present invention relates to the detection of neurological and other movement related disorders.

[2] The invention has been developed primarily for use in the diagnostic field in the determination of neurological diseases which exhibit symptoms related to impaired hand movements among other things. However, it will be appreciated that the invention is not limited to this particular fie Id of use.

Background

[3] P resently, there is a need for a means of reliably determining the presence of movement related neurological disease in a patient with a greater level of accuracy than is currently available. F urthermore, it is vital to the health outcome of a patient that such diseases are picked up in the early stages of progression, often when symptoms are subtle and ha rd to detect. Typica lly, the earlier one ca n detect these neurologica l diseases the earlier treatment may be commenced s lowing the progression of the disease and preventing further neurologica l damage from occurring ^" resulting in significantly improved health outcomes for the patient. C urrently, a specialised physician with a high level of training is required to make a n accurate diagnosis of movement related disorders. Non-specialist clinicians and G eneral P ractitioners (G P s) are generally not qua lified to determine the presence of such disorders with any high level of certainly. Therefore, there is a need for a system that clinicians and primary hea lth care professionals may use to accurately determine the likelihood of the presence of neurological disease.

[4] It is to be understood that, if a ny prior a rt information is referred to herein, such reference does not constitute an admission that the information forms pa rt of the common general knowledge in the art, in Austra lia or any other country.

S ummary

[5] The present disclosure details a system for the reliable detection of neurological diseases exhibiting movement related symptoms. The system involves the analysis of data generated from data entry devices, which may include a keyboa rd, a touch pad or any other human computer interaction (HCI) device used for entering data. To extract the data derived from the patient, a keyboard or other HCI device, or a dedicated device may be used to determine characteristics which exist in patients suffering from neurologica l disease, such characteristics may include; asymmetry related to keyboard use, reaction speed, keystroke hold times, keystroke latency times between sequential keys, jerkiness of motion a nd cha nges throughout the day, or during a predefined time period, inter a lia.

[6] Pa rticula r data processing algorithms may be used in order to provide a technical solution to the technical problem of the determination of neurological disease in patients in the early stages of disease presence, where disease symptoms may be subtle a nd hard to detect. With diseases such as Pa rkinson s disease, it is vitally important in gaining an early diagnosis, as the ea rlier the disease is detected the greater the a bility to impede its progression, a nd the greater the outcome for the patient.

[7] According to an as pect of the present invention, there is provided a system for determining the presence of a movement-related disease in a patient, the system compris ing:

a processor;

a memory storage device comprising computer program code and being operably connected to the processor;

an input device opera bly connected to the processor and adapted to send a n input data to the processor when the input device is used by the patient to input data; and wherein, the computer program code is configured to:

identify patient keystroke feature data in accordance with at least the input data, comprising data corresponding to the following list of patient keystroke timing characteristics:

keystroke hold times (an indication of vertica l movement) and their statistical features; latency between successive key strokes (a n indication of latera l movement) and their statistical features; and

sepa ration of, and differences between left and right hand activity;

for determining the likelihood of the presence of the movement-related disease in accordance with the patient keystroke feature data.

[8] P referably, the patient keystroke feature data is pre-processed using at least one of: linear discriminant analysis;

kernel extreme learning; and machine and subtractive clustering features weighting;

prior to the determination of the likelihood of the presence of the movement-related disease.

[9] P referably, the patient keystroke feature data is pre-processed using a combination of: linear discriminant analysis;

kernel extreme learning; and

machine and subtractive clustering features weighting;

[10] P referably, one or more machine learning models, typically such as a support vector machine (SVM), multi level perceptron, logistic regression model, random forest classifier, nu-support vector classification, decision tree classifier, K-nearest neighbours (KNN) and quadratic discriminant ana lysis, a re applied to the patient keystroke feature data, either individually or after grouping, to determine the likelihood of the presence of the movement-related disease.

[1 1] P referably, the computer program code is further configured to predict the presence of hand tremors by executing at least one of the following list of steps:

comparing the first input data with known input data corresponding to hand tremors using a machine learning matching algorithm;

generating a continuous digita l data stream from the patient keystroke feature data by removing non-contiguous instances;

a pplying pre-processing to the digital data stream by using a low pass signal filter and/or polynomial interpolation to generate a filtered data stream;

a pplying Fourier analysis to the filtered data stream in order to produce a frequency domain representation and identification of pa rticula r frequency components including their respective power levels indicative of hand tremors;

creating a series of non-integer sub-sampling decimations from the digital data stream; and

a pplying an ana lysis of variance to the digital data stream in orderto identify pa rticular frequency and/or power components indicative of hand tremors. [12] According to a further as pect of the pres ent invention, there is provided a system for determining the presence of a movement-related dis eas e in a patient, the system compris ing:

a processor;

a device comprising a tri-axia l accelerometer, which is adapted to be held by or attached to the patient, and which is configured to measure tremors based on the output of the tri-axial accelerometer to produce tremor data and to send the tremor data to the processor;

wherein, the computer program code is configured to determine the likelihood of the presence of the movement-related disease in the patient in accorda nce with the tremor data.

[13] P referably, one or more machine learning models are applied to the tremor data either individua lly or after grouping, to determine the likelihood of the presence of a movement-related disease in the patient in accordance with a presence of two separate cardinal symptoms.

[14] According to a further aspect of the present invention, there is provided a system for determining the presence of Pa rkinson s Disease in a patient, the system comprising: a processor;

an input device operably connected to the processor and adapted to send a first input data to the processor when the input device is used by the patient to input data; and wherein, the computer program code is configured to identify patient keystroke feature data indicative of the presence of Pa rkinson s Disease in accordance with at least the first input data, wherein, the patient keystroke feature data comprises at least an asymmetry of the patients hand movement between left and right-hand fingers;

to determine the likelihood of the presence of Parkinson s disease in accordance with the patient keystroke feature data.

[15] P referably, the patient keystroke feature data is pre-processed using at least one of: linear discriminant analysis; kernel extreme learning; and

machine and subtractive clustering features weighting;

[16] P referably, the patient keystroke feature data is pre-processed using a combination of: linear discriminant analysis;

kernel extreme learning; and

machine and subtractive clustering features weighting;

[17] P referably, the patient keystroke feature data is pre-processed using a hybrid of: linear discriminant analysis;

kernel extreme learning; and

machine and subtractive clustering features weighting;

[18] P referably, one or more machine learning models are applied to the patient keystroke feature data, either individually or after grouping to determine the likelihood of the presence of Parkinson s disease in a patient.

[19] According to a further aspect of the present invention, there is provided a method for determining the presence of Parkinson s disease in a patient, the method comprising the following steps:

receiving patient keystroke feature data;

identifying asymmetry data within the patient keystroke feature data associated with an asymmetry of the patients hand movement; and

determining the likelihood of the presence of Parkinson s disease in accordance with the asymmetry data.

[20] P referably, the patient keystroke feature data is pre-processed using one, a combination, or hybrid of linear discriminant a nalysis, kernel extreme learning machine and subtractive clustering features weighting prior to the determination of the likelihood of the presence of the movement-related disease. [21] P referably, the patient keystroke feature data is grouped with machine lea rning applied at least at an individua l or a group level.

[22] P referably, a plura lity of separate machine learning models, typica lly such as a support vector machine (SVM), multi level perceptron, logistic regression model, random forest classifier, nu-support vector classification, decision tree classifier, K-nearest neighbours (KNN) and quadratic discriminant analysis, is used to create a weighted meta -classifier.

[23] P referably, the computer program code is further configured to predict the presence of the movement-related disease wherein, a hierarchy of prediction models is used to predict movement-related disease severity.

[24] P referably, the method comprises at least one of the following steps:

a. a pplying machine learning in order to compare the patient keystroke feature data with known input data corres ponding to hand tremors;

b. creating a continuous digital data stream from the patient keystroke feature data by removing non-contiguous insta nces;

c. a pplying pre-processing to the continuous digital data stream by using a low pass signal filter and/or polynomial interpolation;

d. a pplying Fourier a na lysis to the continuous digital data stream in order to produce a frequency domain representation and identification of particula r frequency components and their respective power levels indicative of hand tremors;

e. creating a series of non-integer sub-sampling decimations; and

f. a pplying an analysis of variance to the continuous digital data stream in order to identify particula r frequency a nd/or power components indicative of hand tremors.

[25] P referably, a plura lity of separate machine learning models, typica lly such as a support vector machine (SVM), multi level perceptron, logistic regression model, random forest classifier, nu-support vector classification, decision tree classifier, K-nearest neighbours (KNN) and quadratic discriminant analysis, is used to create a weighted meta -classifier, to determine the likelihood of the presence of a movement-related disease in the patient in accorda nce with the presence of two separate cardina l symptoms. [26] P referably, a hiera rchy of prediction models is used to predict the movement- re la ted disease severity.

[27] In a fu rther as pect of the present invention, there is provided a method for determining the presence of a movement-related disease in a patient, the method compris ing the following steps :

a. receiving patient keystroke feature data;

b. identifying the patient keystroke feature data indicative of the presence of the movement-related disease wherein, the patient keystroke feature data comprises data corresponding to a combination of the following list of patient keystroke features:

c. key stroke hold times;

d. latency between successive key strokes; or

e. separation of, and differences between left and right hand activity;

f. for determining the likelihood of the presence of the movement- related disease in accordance with the patient keystroke feature data.

[28] P referably, the patient keystroke feature data is pre-processed using one, a combination, or hybrid of linear discriminant analysis, kernel extreme learning machine a nd subtractive clustering features weighting prior to the determination of the likelihood of the presence of the movement-related disease.

[29] P referably, a plura lity of separate machine learning models, typica lly such as a support vector machine (SVM), multi level perceptron, logistic regression model, random forest classifier, nu-support vector classification, decision tree classifier, K-nearest neighbours (KNN) a nd quadratic discriminant ana lysis, is used to create a weighted meta-classifier.

[30] P referably, the computer program code is further configured to predict the presence of the movement-related disease wherein, a hierarchy of prediction models is used to predict movement-related disease severity.

[31] P referably the method further comprises one or more steps from the following list of steps:

comparing the keystroke feature data with known input data corresponding to ha nd tremors using a machine learning matching a lgorithm;

generating a continuous digita l data stream from the keystroke feature data by removing non-contiguous insta nces; applying pre-processing to the continuous digita l data stream by using a low pass signa l filter and/or polynomial interpolation to generate a first filtered data stream; applying Fourier ana lysis to the first filtered data stream in order to produce a frequency domain representation and identification of particular frequency components and their respective power levels indicative of hand tremors;

applying a n analysis of variance to the digita l data stream in orderto identify particula r frequency and/or power components indicative of hand tremors.

[32] P referably, a plura lity of separate machine learning models, typica lly such as a support vector machine (SVM), multi level perceptron, logistic regression model, random forest classifier, nu-support vector classification, decision tree classifier, K-nearest neighbours (KNN) a nd quadratic discriminant ana lysis, is used to create a weighted meta-classifier, to determine the likelihood of the presence of a movement- re I a ted disease in the patient in accorda nce with the presence of two sepa rate ca rdina l symptoms.

[33] P referably, a hiera rchy of prediction models is used to predict the movement-related disease severity.

[34] According to a further as pect of the present invention, there is provided a method for determining the presence of a movement-related disease in a patient, the method compris ing the fol lowing steps :

receiving an output from a tri-axial accelerometer, which is adapted to be held by or attached to the patient, and which is configured to measure ha nd tremors based on the output of the tri-axial accelerometerto produce tremor data and to send the tremor data to a processor;

inputting the tremor data into the processor;

determining the presence of hand tremors based on the tremor data received by the processor, wherein the processor is configured to determine the presence of ha nd tremors based on the received tremor data;

for determining the likelihood of the presence of a movement-related disease in a patient in accordance with the tremor data.

[35] P referably, a plura lity of separate machine learning models, typica lly such as a support vector machine (SVM), multi level perceptron, logistic regression model, random forest classifier, nu-support vector classification, decision tree classifier, K-nearest neighbours (KNN) and quadratic discriminant analysis, is used to create a weighted meta -classifier, to determine the likelihood of the presence of a movement-related disease in the patient in accorda nce with the presence of two separate cardina l symptoms.

[36] P referably, a hiera rchy of prediction models is used to predict the movement-related disease severity.

[37] According to a fu rther as pect of the present invention, there is provided a method for determining the presence of a movement-related disease in a patient, the method compris ing:

receiving patient keystroke feature data from a n input device;

determining the likelihood of the presence of the movement-related disease based on the ana lysis of the patient keystroke feature data comprising at least one of: key stroke hold times;

latency between successive key strokes; and

sepa ration of, and differences between left and right hand activity;

for determining the likelihood of the presence of the movement-related disease in accordance with the determination based on the patient keystroke feature data.

[38] In another as pect, the invention may be said to cons ist in a method for determining the pres ence of a movement-related dis ease in a patient, the method compris ing the steps of:

a. receiving contiguous patient keystroke data elements from at least one or more input devices;

b. generating a plurality of signal a rrays from the patient keystroke data; c. performing a regression a nalysis on the plurality of signal arrays for a ra nge of possible tremor frequencies at small frequency increments to generate a plurality of s ignal analysis a rrays from each signal array, each signal a na lys is a rray including at least one signa l a nalysis array element; d. ca lculating interpolated values for each s igna l a nalysis array element of each signa l a na lysis array for the particular frequency and elapsed time of that signal analysis a rray, using at least one of sinusoidal or polynomial interpolation; e. determining the difference between the va lue of each signa l ana lysis a rray element and the ca lculated interpolated value for that element as a va riance; a nd

f. compa ring the varia nce of each signa l ana lysis array element at a particular target tremor frequency with the va ria nce of signal analysis a rray elements at other target tremor frequencies to detect one or more selected from i. one or more dominant tremor frequencies a nd

ii. one or more dominant tremor frequency ranges, that are indicative of hand tremor in the patient.

[39] In one embodiment, the patient keystroke data elements correspond to a pa rticular hand of a subject.

[40] In one embodiment, each signal array represents a digital data stream with a nonuniform sampling interva l.

[41] In one embodiment, the wherein each patient keystroke data element comprises at least one or more selected from

a. a n elapsed time from an index time;

b. flight time; and

c. hold time.

[42] In one embodiment, the number of patient keystroke data elements are sufficient for the identification of any component ha nd tremor at particula r frequencies.

[43] In one embodiment, each signa l a nalysis array applies to a particular target tremor frequency and phase angle.

[44] In one embodiment, the step of comparing the variance also includes the step of filtering the signal ana lysis arrays.

[45] In one embodiment, the method comprises the step of compa ring dominant tremor frequencies in one hand of the patient to a nother ha nd.

[46] In one embodiment, the method comprises the step of comparing the phase angle of each detected dominant tremor frequency to proximate frequencies. [47] According to a further as pect of the pres ent invention, there is provided a system for determining the presence of a movement-related disease in a patient, the system compris ing:

a. a processor;

b. a memory storage device comprising computer program code a nd being opera bly connected to the processor;

c. a n input device operably connected to the processor and adapted to send an input data to the processor when the input device is used by the patient to input data;

d. wherein the computer program code is configured for directing the system for: i. receiving patient keystroke feature data;

ii. identifying diagnosis data from the patient keystroke feature data, including at least:

1. flight time data distribution patterns relating to time periods between keystrokes within the patient keystroke feature data for each hand of the patient; and

2. hold time data distribution patterns relating to time periods over which a keystroke is performed within the patient keystroke feature data for each hand of the patient; and

iii. comparing the identified flight time data distribution patterns and the hold time data distribution patterns for each hand to determine asymmetry data; and

iv. compa ring the diagnosis data to similar control data of healthy

persons to determine the likelihood of the presence of Parkinson s disease in accordance with the asymmetry data.

[48] In one embodiment a plurality of machine learning models are a pplied in para llel to the patient keystroke feature data, either individually or after grouping of related features to determine the likelihood of a presence of the movement- re la ted disease, wherein results of each of the machine learning models are combined in a weighted meta- classifier.

[49] In one embodiment statistical attributes of the patient keystroke feature data are pre- processed and dimensionally reduced using at least one of:

a. linear discrimina nt a nalysis;

b. kernel extreme lea rning; or

c. machine and subtractive clustering features weighting; prior to the determination of the likelihood of the presence of the movement- related disease.

[50] In one embodiment the patient keystroke feature data is pre-processed using a combination of:

a. linear discrimina nt a nalysis;

b. kernel extreme lea rning; and

c. machine and subtractive clustering features weighting; prior to the determination of the likelihood of the presence of the movement-related disease.

[51] In one embodiment the computer program code is further configured to predict the presence of hand tremors by executing at least one of the following list of steps: a. compa ring the input data with known statistical input data corresponding to hand tremors using a machine lea rning matching a lgorithm;

b. generating a continuous digital data stream from the patient keystroke feature data by removing non-contiguous insta nces;

c. a pplying pre-processing to the continuous digita l data stream by using a low pass signal filter and/or polynomial interpolation to generate a filtered data stream;

d. a pplying Fourier ana lysis to the filtered data stream in order to produce a frequency domain representation and identification of particula r frequency components including their respective power levels indicative of hand tremors;

e. creating a series of non-integer sub-sampling decimations from the digital data stream; applying a n a nalysis of variance to the digital data stream in order to identify particular frequency and/or power components indicative of hand tremors;

f. a pplying a method comprising the steps of

i. receiving contiguous patient keystroke data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke data elements defining feature va lues with associated time stamps;

ii. generating at least one or more data arrays from the patient keystroke data elements; iii. for each data array, determine the differences between the feature value of a data element and an associated value of a predetermined wave form at a particula r frequency a nd time stamp, over a range of possible tremor frequencies at a plurality of frequency increments, to generate an analysis array;

iv. calculating the variance of the determined differences for each

frequency increment; or

v. compa ring the varia nce of the particula r frequency to a predetermined threshold generated from the varia nces of the frequencies in the frequency ra nge to identify one or more candidate tremor frequencies.

[52] According to a further as pect of the pres ent invention, there is provided a method for determining the pres ence of Parkins on s disease in a patient, the method comprising the following steps :

a. receiving patient keystroke feature data;

b. identifying diagnosis data from the patient keystroke feature data, including at least:

i. flight time data distribution patterns relating to time periods between keystrokes within the patient keystroke feature data for each hand of the patient; and

ii. hold time data distribution patterns relating to time periods over which a keystroke is performed within the patient keystroke feature data for each hand of the patient; and

c. compa ring the identified flight time data distribution patterns and the hold time data distribution patterns for each ha nd to determine asymmetry data; and d. compa ring the diagnosis data to simila r control data of healthy persons to determine the likelihood of the presence of Parkinson s disease in accordance with the asymmetry data.

[53] In one embodiment the method comprises the step of identifying a plurality of flight time distribution patterns and a plurality of hold time distribution patterns.

[54] In one embodiment the method comprises the step of using the patient keystroke feature data to determine the presence of tremors in a patient.

[55] In one embodiment the patient keystroke feature data is pre-processed using at least one of:

a. linear discrimina nt a nalysis;

b. kernel extreme lea rning machine; or c. subtractive clustering features weighting;

d. prior to the determination of the likelihood of the presence of the movement- related disease.

[56] In one embodiment the patient keystroke feature data is grouped with machine learning applied at least at a n individual or a group level.

[57] In one embodiment a plurality of separate machine lea rning models is used to create a weighted meta-classifier.

[58] In one embodiment a hierarchy of prediction models is used to predict movement- related disease severity.

[59] In one embodiment there exists a method comprising at least one of the following steps:

a. a pplying machine lea rning in order to compare the patient keystroke feature data with known input data corres ponding to hand tremors;

b. creating a continuous digita l data strea m from the patient keystroke feature data by removing non-contiguous insta nces;

c. a pplying pre-processing to the continuous digita l data stream by using a low pass signal filter and/or polynomial interpolation;

d. a pplying Fourier ana lysis to the continuous digita l data stream in order to produce a frequency domain representation a nd identification of particula r frequency components and their respective power levels indicative of ha nd tremors;

e. creating a series of non-integer sub-sampling decimations;

f. a pplying a n ana lysis of va riance to the continuous digital data stream in order to identify particular frequency and/or power components indicative of ha nd tremors;

g. a method comprising the steps of

ii. generating at least one or more data arrays from the patient keystroke data elements;

iii. for each data array, determine the differences between the feature value of a data element and an associated value of a predetermined wave form at a particula r frequency a nd time stamp, over a range of possible tremor frequencies at a plurality of frequency increments, to generate an analysis array;

iv. calculating the variance of the determined differences for each

frequency increment; or

[60] In one embodiment a plurality of separate machine lea rning models is used to create a weighted meta-classifier, to determine the likelihood of the presence of a movement- related disease in the patient in accordance with the presence of at least two separate cardinal symptoms.

[61] In one embodiment a hiera rchy of prediction models is used to predict the movement- related disease severity.

[62] According to a further as pect of the pres ent invention, there is provided a method for determi ning the presence of hand tremor indicative of a movement^" related disease in a patient, the method compris ing the steps of:

a. receiving contiguous patient keystroke data elements from at least one or more input devices for at least one ha nd of a patient, the patient keystroke data elements defining feature values with associated time stamps; b. generating at least one or more data arrays from the patient keystroke data elements;

c. for each data array, determining the differences between the feature value of a data element and an associated value of a predetermined wave form at a particular frequency and time stamp, over a ra nge of possible tremor frequencies at a plurality of frequency increments, to generate an ana lysis a rray;

d. calculating the va riance of the determined differences for each frequency increment;

e. compa ring the variance of the particula r frequency to a predetermined

threshold generated from the varia nces of the frequencies in the frequency range to identify one or more candidate tremor frequencies.

[63] In one embodiment the predetermined threshold is a predetermined proportion of the mea n of the variances.

[64] In one embodiment the method further comprises the steps of: a. for at least the one or more ca ndidate tremor frequencies and a

predetermined range of proximate frequencies, determine a best fit of a predetermined wave form against the feature va lues for a range of phase a ngles to identify the phase a ngle of the ca ndidate frequency and proximate frequencies; and

b. compa ring the identified phase a ngles of the at least one or more candidate frequencies with the identified phase angles of the proximate frequencies in order to confirm candidate frequencies as confirmed frequencies.

[65] In one embodiment there exists a method comprising the step of:

a. compa ring confirmed frequencies of the data a rray to the confirmed

frequencies of another data array.

[66] In one embodiment there exists a method further comprising the steps of:

a. determining the sum of the differences between the va ria nces of the

candidate frequencies and a predetermined threshold relative to va ria nces at other frequencies;

b. determining the number of candidate frequencies;

c. diagnosing the presence of a movement related disease based on a

combination of the determined sum of the differences and the number of candidate frequencies.

d. replicate process with new dataset to confirm frequencies for patient.

[67] According to a further as pect of the present invention, there is provided a system for determining the presence of a movement-related disease in a patient, the system compris ing:

a. a processor;

d. wherein the computer program code is configured for directing the system for: i. receiving contiguous patient keystroke data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke data elements defining feature va lues with associated time stamps;

ii. generating at least one or more data arrays from the patient keystroke data elements; iii. for each data array, determining the differences between the feature value of a data element and an associated value of a predetermined wave form at a particula r frequency a nd time stamp, over a range of possible tremor frequencies at a plurality of frequency increments, to generate an analysis array;

iv. calculating the variance of the determined differences for each

frequency increment;

[68] In one embodiment the predetermined threshold is a predetermined proportion of the mea n of the variances.

[69] In one embodiment the computer program code is configured for directing the system for:

a. for at least the one or more ca ndidate tremor frequencies and a

[70] In one embodiment the computer program code is configured for directing the system for:

a. compa ring confirmed frequencies of the data a rray to the confirmed

frequencies of another data array.

[71] In one embodiment the computer program code is configured for directing the system for:

a. determining the sum of the differences between the va ria nces of the

b. determining the number of candidate frequencies;

c. diagnosing the presence of a movement related disease based on a

combination of the determined sum of the differences and the number of candidate frequencies. d. replicate process with new dataset to confirm frequencies for patient.

[72] According to a further as pect of the pres ent invention, there is provided a method of diagnosi ng Parkins on s disease compris ing the steps of:

a. receiving contiguous patient keystroke feature data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke feature data elements defining keystroke feature va lues with associated time stamps; a nd

b. processing the keystroke feature values to detect both bradykinesia and

tremor in a patient.

[73] According to a further as pect of the pres ent invention, there is provided a system for determining the presence of a movement-related disease in a patient, the system compris ing:

a. a processor;

b. a memory storage device comprising computer program code a nd being

opera bly connected to the processor;

d. wherein the computer program code is configured for directing the system for: i. receiving contiguous patient keystroke feature data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke feature data elements defining keystroke feature values with associated time stamps;

ii. processing the keystroke feature va lues to detect both bradykinesia a nd tremor in a patient.

[74] According to a further as pect of the pres ent invention, there is provided a method for determining the pres ence of a movement-related diseas e in a patient, the method compris ing the steps of:

b. generating a plura lity of signa l a rrays from the patient keystroke data;

c. performing a regression analysis on the plura lity of signa l arrays for a range of possible tremor frequencies at sma ll frequency increments to generate a plurality of signa l analysis arrays from each signa l a rray, each signal a nalysis a rray including at least one signal analysis array element; d. calculating interpolated values for each signa l analysis array element of each signa l analysis array for the particula r frequency and elapsed time of that signa l analysis array, using at least one of sinusoidal or polynomial interpolation;

e. determining the difference between the value of each signa l ana lysis array element and the calculated interpolated value for that element as a variance; a nd

f. compa ring the variance of each signal ana lysis array element at a particula r target tremor frequency with the variance of signal analysis a rray elements at other target tremor frequencies to detect one or more selected from i. one or more dominant tremor frequencies a nd

ii. one or more dominant tremor frequency ranges, that a re indicative of hand tremor in the patient.

[75] According to a further as pect of the pres ent invention, there is provided a system for determining the presence of a movement-related disease in a patient, the system compris ing:

a. a processor;

d. wherein the computer program code is configured for directing the system for: i. receiving contiguous patient keystroke data elements from at least one or more input devices;

ii. generating a plurality of signal arrays from the patient keystroke data; iii. performing a regression ana lysis on the plura lity of signa l arrays for a range of possible tremor frequencies at sma ll frequency increments to generate a plurality of signal analysis arrays from each signa l array, each signal a na lysis a rray including at least one signa l a na lysis array element;

iv. calculating interpolated values for each signal analysis a rray element of each signal a nalysis array for the particular frequency and elapsed time of that signal a nalysis array, using at least one of sinusoidal or polynomial interpolation; v. determining the difference between the value of each signal ana lysis a rray element and the calculated interpolated value for that element as a varia nce; and

vi. compa ring the varia nce of each signal analysis array element at a particular target tremor frequency with the variance of signa l analysis a rray elements at other target tremor frequencies to detect one or more selected from

1. one or more domina nt tremor frequencies, and

2. one or more domina nt tremor frequency ra nges, that are indicative of hand tremor in the patient.

[76] In another as pect, the invention may be said to cons ist i n a method for determining the pres ence of ha nd tremor i ndicative of a movement^" related diseas e in a patient, the method compris ing the steps of:

a. receiving contiguous patient keystroke data elements from at least one or more input devices for at least one ha nd of a patient, the patient keystroke data elements defining feature values with associated time stamps;

b. generating at least one or more data arrays from the patient keystroke data elements;

c. a pplying regression analysis of a sinusoidal wave to each data array for a range of incrementa l frequencies, over a range of phase angles for each frequency, to establis h which of the frequency increments provides a best fit to the data array to establish one or more dominant frequencies.

[77] T his invention may a lso be said broadly to consist in the parts, elements and features referred to or indicated in the specification of the application, individually or collectively, and a ny or all combinations of any two or more of said parts, elements or features, and where specific integers are mentioned herein which have known equiva lents in the art to which this invention relates, such known equivalents are deemed to be incorporated herein as if individua lly set forth.

[78] Other aspects of the invention are also disclosed.

B rief Des cription of the Drawings

[79] Notwithstanding any other forms which may fall within the scope of the present invention, a preferred embodiment of the invention will now be described, by way of example only, with reference to the accompanying drawings in which: [80] F ig. 1 shows one embodiment of an input device ;

[81] F ig. 2 shows the attachment of a tri-axial accelerometer to a patients hand for the detection of tremors;

[82] F ig. 3 shows the output of the tri-axia l accelerometer for use in the determination of disease presence;

[83] F ig. 4 shows a flow chart of a first embodiment of a method for determining Parkinson s disease using patient keystroke data;

[84] F ig. 5 shows a flow cha rt of a second embodiment of a method for determining Parkinson s disease using patient keystroke data;

[85] F ig. 6 shows a flow chart for a method of detecting hand tremors from a n input device; and

[86] F ig. 7 shows a flow chart for a method of determining of movement related disease using tremor data.

Des c ri ption of E mbod i ments

[87] It should be noted in the following description that like or the same reference numera ls in different embodiments denote the same or similar features.

[88] F igures 1 , 2 and 3 exemplify the use of the device for the detection of movement- related disorders. In Figu re 1 the system 100 comprises a n input device, which in this embodiment is a keyboard 104 opera bly connected to a computer 102 consisting of a processor 106, memory 108 and output 1 10. The patient inputs data into the computer 102 via the keyboa rd 104 where the keyboard 104 is configured to extract data specific to the patient s hand 1 12 movements. The extracted data is then fed into the computer for processing in order to determine the likelihood of disease presence in the patient. The processing methods will be discussed in detail later.

[89] In F igure 2 the attachment of a tri-axia l accelerometer 202 is shown attached to the patient s hand 1 12 which in this embodiment is by way of a clamp device for attachment to the patient s hand. The tri-axia l accelerometer 202 outputs frequency related data via a signa l transmission means, which in this embodiment is a cable 204 operably connected to the computer 102 of F igure 1 to aid in the determination of disease presence. The tri-axial accelerometer 202 is one embodiment of a dedicated device configured to extract data s pecific to the patient s hand movements. It is understood that any suita ble device to extract data specific to the patient s ha nd movements is within the scope of this discussion. [90] F igure 3 exemplifies the process of ana lysing output data from the tri-axia l accelerometer 202 by the attachment of the tri-axia l accelerometer 202 to the patient s hand 1 12. The tri-axia l accelerometer 202 then generates a signal in accordance with the movement of the patient s hand 1 12, which is sent via a cable 204 to the computer 102 in this embodiment.

[91] In F igu re 3 the output of the tri-axial accelerometer 202 may be processed by performing a Fourier transform (which may be performed by applying a F FT) on the signa l output of the tri-axial accelerometer 202 as is exemplified in F igure 3 as the output frequency spectra chart. Hand tremors related to movement-related disorders typically exhibit unique frequency spectra, with output frequencies in the range of 4-6 Hz being common in disorders such as Parkinson s Disease. In combination with the extraction of keyboard 104 entry data (to be discussed in detail later) correlated with movement-related disorders, the detection of hand tremors characteristic of movement- related disease can aid in improving the accuracy of diagnosis a nd reduce the risk of false positive diagnosis.

[92] F igures 4, 5 and 6 show similar processes for providing output for use in the determination of Parkinson s disease and other movement-related diseases. In Figure 4, after receiving patient keystroke data, features from the patient keystroke data are identified, which may include data suggestive of bradykinesia (which is characterised by slow or erratic keystrokes or delays in repetitive keystroke input) a nd akinesia (which is related to slowness in initiating movement) or tremors, all of which are cha racteristic of some movement-related diseases which may be detected by the analysis of keyboa rd input data as produced by a patient, one such method will be discussed in detail in relation to F igure 6. F urthermore, an electro-mechanica l method of tremor detection will be discussed in relation to Figure 7 below. In regards to patient keystroke feature data, other features that can be extracted from the patient keystroke data include data related to asymmetry or sidedness characteristics related to the patient s hand movements among other possible characteristics may be extracted from the keystroke feature data.

[93] Two optional methods are shown in Figu re 4, one in which the keystroke features are pre-processed, using linea r discriminant ana lysis, kernel extreme learning or machine and/or subtractive clustering features weighting to name a few possible methods. In another embodiment, the keystroke features are grouped after pre-processing into those involving vertical finger movement a nd those involving lateral finger movement. Once the keystroke features have been grouped, a plura lity of machine lea rning models, typically such as a support vector machine (SVM), multilevel perceptron, logistic regression model, random forest classifier, nu-support vector classification, decision tree classifier, K-nearest neighbours (KNN) a nd quadratic discriminant ana lysis, is then used to create a meta-classifier, which in turn is used to determine the severity of disease if it has been determined that the patient has a movement-related disease.

[94] As shown in Figure 4 a further technique may be used after receiving the patient keystroke data which additiona lly provides detection of hand tremors. Hand tremors, typically occurring at a frequency between 4-6 Hz in one or both hands, a re often associated with movement-related diseases and their detection provides a va luable mea ns in the diagnosis of such diseases a nd provides a second diagnostic ca rdina l feature of Pa rkinson s Disease. T he technique shown in Figure 4 involves creating a continuous digital data stream from the keystroke data and applying pre-processing to the digital data stream such that it is in a form to perform frequency spectra a nalysis. One embodiment uses the a pplication of Fourier ana lysis to the pre-processed digita l data stream which originated from the input device. In another embodiment a vector representation of the data stream is created where, for a range of frequencies from 3 to 12 Hz, each element comprises the difference between the sampled value of flight time and/or hold time a nd the corresponding sinusoida l interpolated value at the same elapsed time and frequency; and then a n a na lys is of varia nce is applied to each of those vectors in order to identify the frequency spectra.

[95] Based on the ana lysis of the frequency spectra a determination may be made as to the determination of tremors which a re indicative of a movement related disease.

[96] Asymmetry of ha nd movements, particula rly in the early stages of Parkinson s disease, is a characteristic which is detectable and may provide a valuable pa rameter in aiding in the determination of a positive disease diagnosis. By sepa rately a nalysing the left and right-ha nded keystroke data, followed by the application of machine learning and other techniques, which will be discussed in more detail later, such asymmetry or sidedness cha racteristics related to ha nd movements may be detected. In addition, the keystroke features from just the most affected side will provide a more accurate determination of the presence of Parkinson s disease. A further novel method of detecting hand tremor using input device data will be discussed in relation to Figure 6.

[97] In Figu re 5, a flow cha rt outlines another embodiment of the invention, which targets the detection of Parkinson s disease using asymmetry of movement, in particular, involving the steps of receiving patient keystroke data from an input device which may be a ny huma n computer interface device, such a keyboard or touch pad, where a person interacts with the device in order to enter data. Once the patient keystroke feature data is received each keystroke is categorised as being either left-ha nded or right-ha nded depending on its particula r position with respect to the keyboard layout. S ome keystrokes are excluded, for example those such as the space key and characters in the centremost section of the keyboard, which may be typed by fingers of either hand. The qua lity of the data derived from the input device may be improved by performing data pre-processing on the patient keyboard input features. In one embodiment, the keystroke features are grouped after pre-processing into those involving vertical movement and those involving lateral movement, Then a plurality of machine learning models is applied sepa rately to the left a nd right ha nd features to create two meta-classifiers, the results of which are used to determine whether there is asymmetry and the likelihood of the presence of a movement-related disease such as Parkinson s Disease. In turn this is used to determine the severity of disease if it has been determined that the patient has a movement-related disease.

[98] With reference to Figure 6, a method for determining ha nd tremors using input data, which may be in the form of input data from a keyboard, touchscreen input data, or any other a ppropriate device that a user may use to input information, particula rly in a contiguous form of a number of data points in the vicinity of 50 or more inputs for example.

[99] The method to be described below makes use of a technique to be termed ivariable frequency sinusoida l interpolation ^"the details of which will be discussed later in the specification.

[100] Referring again to Figure 6, we start with receiving 701 contiguous input data from an input device, as discussed previous ly the input device may be a keyboa rd, mobile phone touchscreen, tablet or any other device in which the user is likely to input a reasonable level of data in the form of keystrokes for instance. Once the data has been input, it is captured through the generation 703 of a plurality of data arrays, which may be of a variable length or dimension depending on the length of the keystroke input interval and the number of input data the user inputs. Next a plurality of signal a nalysis arrays 705, from the data analysis a rrays. The signal analysis arrays are used to store data particula r to the signals cha racteristics. A set of interpolated va lues are then generated 707 at each ta rget frequency from which a difference between each component of the signal ana lysis array is ca lculated 709. Filtering 71 1 is then performed on the signa l analysis array components followed by a determination 713 of the variance of each of the signal ana lysis array components is taken. From which a determination of the likelihood of the presence of hand tremors is determined 71 5 by determining the target frequencies with the lowest varia nce.

[101] In Figu re 7, the determination of hand tremors is performed using an external apparatus for the detection of hand tremors. A tri-axial accelerometer is one such means of producing accurate movement related data for use in the determination of hand tremors and, by a pplying Fourier analysis, the frequency spectra of the output of the tri-axia l accelerometer can be a na lysed to look for frequency components indicative of hand tremors. In this embodiment, the external apparatus is a tri-axial accelerometer which is used for the determination of ha nd movements related to movement- re la ted neurological diseases, in pa rticula r for the determination of ha nd tremors as mentioned ea rlier. The tri-axial accelerometer may be attached to a patient s ha nd in order to monitor the motion of the patient s hand for the determination of movement-related disease presence. In this embodiment, the output of the tri-axial accelerometer is analysed to determine whether there exists the presence of ha nd tremors. As with previous embodiments the data received from the tri-axial accelerometer may be pre-processed before being analysed for the detection of tremors, which may aid in the accuracy of the determination of disease presence in the patient. In addition, by combining this with the analysis of patient keystroke data described previously in Figu re 4, a determination of the presence of Pa rkinson s Disease using two separate cardinal features is achieved.

[102] The three embodiments of Figu res 4, 5 and 6 can similarly involve similar sub- processes for providing information related to disease severity in the patient in addition to the aforementioned detection of hand tremors and a nalysis of data derived from HCI devices (such as a keyboard).

[103] The present disclosure details a system for determining the presence of a movement- re la ted disease in a patient. Movement related diseases come in a variety of different forms, with Parkinson s Disease being one of the most well-known a nd one of the more debilitating neurologica l diseases eventually leading to death. In one embodiment, the system for detecting neurological disease may comprise of a processor; a memory storage device comprising computer program code and being opera bly connected to the processor; an input device, which may be a keyboard, touch pad or any other HCI device opera bly connected to the processor and adapted to send an input data to the processor when the input device is used by the patient to input data. T he computer program code may be configured to identify patient keystroke feature data in accordance with at least the input data, the input data may correspond to the following list of patient keystroke features: key stroke hold times;

latency between successive key strokes; and

sepa ration of, and differences between left a nd right hand activity.

[104] The above features are a subset of possible features which may be used to determine the likelihood ofthe presence of the movement-related disease in a patient. The more features that a re used in the diagnosis process, the greater the level of accuracy that may be obtained in determining the likelihood of the presence of a neurological disease. As will be discussed later, other parameters may be used in the determination of neurological disease, in order to improve the accuracy of disease determination.

[105] One of the key features of the system for diagnosis of movement-related disease as disclosed, is the a bility to diagnose with a high level of certainty the presence of movement- re la ted diseases using a sta ndard computer which almost all patients would have. C omputers are typically in widespread use by the demographic that Pa rkinson s Disease and other related neurological diseases tend to affect. Ma ny of these people would use a computer most days of the week, a nd hence would be constantly inputting data into their computer or other related device as a fact of everyday life. A detection methodology which directly utilises such data and which does not require a supervised environment or the typing of fixed passages of text will be of significant benefit.

[106] Many features such as asymmetry of keyboard use and latency between keystrokes have been shown to be readily extracted from the keyboard input data. S uch data ca n be invaluable in the determination of movement-related disease.

[107] In order to improve the accuracy of diagnosis of a plurality of machine lea rning models may be applied to the patient keystroke feature data as was discussed previously, either for individua l features or after grouping of related features to determine the likelihood of the presence of the movement-related disease. Machine learning allows the accuracy of the system for diagnosis of the movement-related disease to improve over time, improving the qua lity of determination output in correlation with the patient s use of the system.

[108] In order to further improve the likelihood of correct diagnosis, the patient keystroke feature data may be pre-processed using at least one of:

linear discrimina nt a na lysis;

kernel extreme learning; and machine and subtractive clustering features weighting.

[109] P re-processing is performed prior to the determination of the likelihood of the presence of the movement-related disease, improving the likelihood of correct disease diagnosis.

[1 10] In addition to pre-processing the patient keystroke feature data using at least one of the abovementioned pre-processing techniques, pre-processing may be performed using a combination (in contrast to at least one of):

linear discrimina nt a na lysis;

kernel extreme learning; and

machine and subtractive clustering features weighting;

[1 1 1] The pre-processing is performed prior to the determination of the likelihood of the presence of the movement-related disease.

[1 12] Fina lly, the patient keystroke feature data may be pre-processed a nd dimensionally reduced using a hybrid of:

linear discrimina nt a na lysis;

kernel extreme learning; and

machine and subtractive clustering features weighting;

[1 13] With the above pre-processing techniques being applied prior to the determination of the likelihood of the presence of the movement-related disease.

[1 14] P refera bly, the pre-processing is a pplied prior to determination as the pre-processing is aimed at improving the quality of the determination data, in order to reduce the likelihood of false positive results.

[1 15] In addition to the aforementioned techniques the computer program code may be further configured to predict the presence of ha nd tremors by executing at least one of the following list of steps:

compa ring the first input data with known input data corresponding to hand tremors using a machine lea rning matching algorithm;

applying pre-processing to the digita l data stream by using a low pass signal filter and/or polynomial interpolation to generate a filtered data stream; applying Fourier analysis to the filtered data stream in order to produce a frequency domain representation and identification of particular frequency components including their respective power levels indicative of hand tremors;

creating vector representations of the data stream where, for a range of frequencies from 3 to 12 Hz, each element comprises the difference between the sampled value of flight time and/or hold time and the corresponding sinusoidal interpolated value at the same elapsed time and frequency; a nd

applying an analysis of variance to each of those vectors in order to identify particula frequency and/or power components indicative of hand tremors.

applying an analysis of variance to the digital data stream in order to identify particular frequency and/or power components indicative of hand tremors.

[1 16] The above steps are related to a different method for the determination of the movement- re la ted disease in the patient. Previously discussed methods involved the determination of disease presence by analysis of keystroke feature data. To further improve diagnosis accuracy one may analyse physical movement data, which in this case involves the detection of hand tremors which are a characteristic of movement related diseases such as Parkinson s Disease inter alia. In the above steps, a keyboard is used as the input device however, as will be discussed laterotherdevices may be used to reliably detect hand tremors. The advantage of the use of a keyboard is that no external apparatus is required meaning a standard computer or other input device is all that is required from a hardware standpoint.

[117] Tremor analysis is a valuable means for determining the presence of movement- related diseases, in particular Parkinson s Disease, which commonly reveals itself with the presence of hand tremors, typically occurring in the rest state in between 70- 75% of patients with Parkinson s Disease. Therefore, the determination of hand tremors specific to Parkinson s disease is an important diagnostic tool in movement- related diseases such as Parkinson s Disease. In one embodiment, the system for detecting hand tremors in a patient for determining the presence of a movement- related disease in a patient, may comprise:

a processor; a memory storage device comprising computer program code a nd being operably connected to the processor; and

a device comprising a tri-axia l accelerometer, which is adapted to be held by or attached to the patient, a nd which is configured to measure tremors based on the output of the tri-axia l accelerometer to produce tremor data and to send the tremor data to the processor.

[1 18] The computer progra m code ca n be configured to determine the likelihood of the presence of the movement-related disease in the patient in accorda nce with the tremor data produced by the tri-axial accelerometer. T he tri-axial accelerometer is one of a number of different devices which may be used in detecting general movements of limbs, other examples may include electromyographic devices a nd computer mouse devices.

[1 19] One or more machine lea rning models a re a pplied to the tremor data either individually or after grouping, to determine the likelihood of the presence of a movement- re la ted disease in the patient in accordance with a presence of two sepa rate ca rdinal symptoms.

[120] S imila rly to the system for determining a movement-related disease a system for determining the presence of Parkinson s Disease in a patient, the system may comprise:

a processor;

a memory storage device comprising computer program code a nd being operably connected to the processor;

an input device opera bly connected to the processor and adapted to send a first input data to the processor when the input device is used by the patient to input data; a nd

[121] The computer program code is configured to identify patient keystroke feature data indicative of the presence of Parkinson s Disease in accordance with at least the first input data, wherein, the patient keystroke feature data comprises at least an asymmetry of the patient s hand movement between left and right-hand fingers in order to determine the likelihood of the presence of Parkinson s Disease in accordance with the patient keystroke feature data.

[122] As in previous methods, the patient keystroke feature data may be pre-processed and dimensionally reduced using at least one of, a combination or a hybrid of:

linear discrimina nt a na lysis; kernel extreme learning; and

machine and subtractive clustering features weighting;

[123] The a bove pre-processing is preferably performed prior to the determination of the likelihood of the presence of the movement- related disease in the patient.

[124] Furthermore, one or more machine learning models a re applied to the patient keystroke feature data, either individually or after grouping to determine the likelihood of the presence of Parkinson s Disease in a patient.

[125] With the method for determining the presence of Parkinson s Disease in a patient, patient keystroke data is a valua ble source of data in making the determination. The method may comprise the following steps:

receiving patient keystroke feature data;

determining the likelihood of the presence of Parkinson s Disease in accorda nce with the asymmetry data.

[126] As in previous methods, the patient keystroke feature data ca n be pre-processed and dimensionally reduced using one, a combination, or hybrid of linear discrimina nt ana lysis, kernel extreme lea rning machine and subtractive clustering features weighting prior to the determination of the likelihood of the presence of the movement- re la ted disease.

[127] Furthermore, the patient keystroke feature data can be grouped with machine learning a pplied at least at an individua l or a group level where a plurality of sepa rate machine learning models ca n be used to create a weighted meta -classifier.

[128] With tremors being a common symptom of movement-related neurological diseases tremor detection is a valuable method to be used in the determination of the disease presence. There are a variety of different devices that ca n be used to detect hand tremors with a tri-axial accelerometer being a reliable a nd sensitive means of detecting tremors however, other devices are within the scope of this discussion. Use a tri-axial accelerometerfor detecting the presence of movement related disease may aid in the accurate determination of the presence of the movement-related disease in addition with the aforementioned methods, the va lue in using a plura lity of different determination methods is that the accuracy of diagnosis tends to increase as the number of determination methods increases. The use of a va riety of different determination means increases the accuracy rate in making such determinations. In one embodiment, a method for determining the presence of a movement-related disease in a patient using an external device for detecting ha nd tremors, may comprise the following steps:

receiving an output from a tri-axial accelerometer, which is adapted to be held by or attached to the patient, and which is configured to measure ha nd tremors based on the output of the tri-axial accelerometer to produce tremor data a nd to send the tremor data to a processor;

inputting the tremor data into the processor;

determining the presence of hand tremors based on the tremor data received by the processor, wherein the processor is configured to determine the presence of ha nd tremors based on the received tremor data for determining the likelihood of the presence of a movement-related disease in a patient in accordance with the tremor data.

Variable F requency S inusoidal Interpolation

[129] In Figu re 6 a method for determining ha nd tremors from an input device was exemplified, the system may be termed Va ria ble F requency S inusoida l Interpolation and will be described in further detail below:

[130] Va riable frequency sinusoida l interpolation offers a novel means of extracting tremor data from a patient s keystroke data. T he first step in this methodology is to extract contiguous sequences of keystrokes from the participant data, with each sequence for each hand comprising an array of ^"

I Keystroke data elements: timestamp (which is the non-uniform sample interval), flight time va lue and hold time va lue. The sepa rate treatment of L and R hands is needed as P D ha nd tremors are always asymmetrical, affecting one side more than the other (if there is a symmetrical tremor, it may indicate E ssentia l T remor rather than P D).

I At least 50 continuous keystrokes, the sequence being terminated by either (i) 120 seconds ela psed time, or (ii) a pause of 10 seconds or more. These constraints are needed, as otherwise any hand tremor signal is likely to be phase-shifted over a longer period of typing, or between non-contiguous sections of typing.

Detection algorithm

[131] The technique uses a sinusoida l interpolation of the contiguous keystroke sequences to test for a particula r tremor frequency, calculating the difference between each sampled flight time value (or, alternatively, hold time) and the corres ponding interpolated value at the same ela psed time. The va riance of a ll those individua l differences will be lower when a tremor is present at that frequency. T hen, by sequentially sca nning across the range of 3 to 10 Hz at sma ll increments, the presence of a ny tremor at a particular frequency, along with its respective phase angle, can be accurately detected.

[132] P referably, there are a total of at least 20 such keystroke sequences used for each pa rticipant, in order to provide replication a nd improve the detection accuracy.

[133] The fina l criteria for determining that an individua l has a P D rest tremor in a pa rticular hand may include ^"

a) T hat the tremor frequency is detected across at least 4 separate replications b) T hat its detection level and strength^" over the 4 to 6 Hz frequency, range exceed a redetermined threshold, calculated as ^"

(1 )

[135] where hand is left or right, f is the frequency (Hz), P^Wfis the mea n amplitude of tremor frequencies (over the 4 to 6 Hz range) for that ha nd and x a nd y a re pre-set threshold values.

Filters

[136] Because of the high detection sensitivity involved and the low S NR of the signa l, it is essential to apply filtering to any :possible ^"tremor frequencies detected. This comprises ^"

1. Consistency

[137] T he width of any detection :notch^" extends over a frequency of a pproximately e 0.1 Hz. This means that, for any detected frequency, if the sta ndard deviations of the immediately-adjacent frequencies are also less than the mea n and, most importantly, their phase angles a re close to that of the detected frequency), the signal ca n be considered as valid, thus eliminating spurious detections.

where f is the frequency (Hz), S DVa lue is the standard deviation of the signal (actual cf. ca lculated) at a pa rticula r frequency, and d is the phase a ngle (degrees). 2. S ide lobe detection

A tremor detection is indicated by both a negative :notch ^" a nd two accompanying positive-value side lobes. The presence of both these lobes can be identified by evaluating the sta ndard deviation va lue at e 0.2 to 0.3 Hz which further reduces any spurious detections.

where f is the frequency (Hz), S DVa lue is the sta ndard deviation of the signal (actual cf. ca lculated) at a pa rticular frequency.

[138] As in previous methods a plura lity of separate machine learning models may be used to create a weighted meta -classifier, to determine the likelihood of the presence of a movement-related disease in the patient in accordance with the presence of two sepa rate ca rdinal symptoms.

[139] F urthermore, a hierarchy of prediction models is used to predict the movement- related disease severity.

[140] T he system as disclosed achieves a high level of diagnostic accuracy through the use of a broad range of keystroke characteristics indicative of the presence of movement related neurological disease. As discussed previously, the features of relevance in the diagnosis of the wide variety of neurological diseases include two- dimensiona l finger movement analysis including features such as vertica l finger hold times, latency between keys, separation between left a nd right hands, a nd their respective statistica l features.

[141] Once keystroke data is extracted from the keyboard data entry, the next component is to process the information in order to derive a result regarding the likelihood of movement related disease in the patient. There are a number of different methods used to process the input data including the dimensionality reduction of features as part of data pre-processing and such methods include but are not limited to linear discriminant analysis, the grouping of features with machine learning a pplied at either the group or individual level, the use of a n ensemble of multiple machine learning models to create a weighted meta-classifier, and a hierarchy of prediction models to predict disease severity. [142] In addition, the machine learning models, once initially trained on patients with known disease status, is saved so thatthe data ca pture a nd determination process may be a pplied to new patients with unknown disease status.

[143] F urthermore, additiona l features, types of features and separate tests may be incorporated into the models a nd the meta classifier.

[144] F inally, the use of machine learning models means that additiona l patient data can be incorporated over time, thus continually improving the detection accuracy.

Interpretation

E mbodiments:

[145] Reference throughout this specification to one embodiment_, or an embodiment_, means that a pa rticular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. T hus, appea rances of the phrases In one embodiment_, or In an embodiment_, in va rious places throughout this specification are not necessarily all referring to the same embodiment, but may. F urthermore, the particular features, structures or characteristics may be combined in any suitable manner, as would be apparent to one of ordinary skill in the art from this disclosure, in one or more embodiments.

[146] S imila rly it should be appreciated that in the above description of example embodiments of the invention, various features of the invention a re sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understa nding of one or more of the various inventive aspects. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed invention requires more features than a re expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less tha n all features of a single foregoing disclosed embodiment. Thus, the claims following the Detailed Description of S pecific E mbodiments a re hereby expressly incorporated into this Detailed Description of S pecific E mbodiments, with each claim standing on its own as a sepa rate embodiment of this invention.

[147] F urthermore, while some embodiments described herein include some but not other features included in other embodiments, combinations of features of different embodiments a re meant to be within the scope of the invention, and form different embodiments, as would be understood by those in the art. For example, in the following claims, any of the claimed embodiments can be used in a ny combination. Different Instances of Objects

[148] As used herein, unless otherwise s pecified the use of the ordinal adjectives 'first., second _, ^'third _, etc., to describe a common object, merely indicate that different insta nces of like objects a re being referred to, and are not intended to imply that the objects so described must be in a given sequence, either temporally, spatially, in ranking, or in any other manner.

S pecific Details

[149] In the description provided herein, numerous specific details are set forth.

However, it is understood that embodiments of the invention may be practiced without these s pecific details. In other instances, well-known methods, structures a nd techniques have not been s hown in detail in order not to obscure an understanding of this description.

Terminology

[150] In describing the preferred embodiment of the invention illustrated in the drawings, specific terminology will be resorted to for the sake of cla rity. However, the invention is not intended to be limited to the s pecific terms so selected, a nd it is to be understood that each s pecific term includes a ll technical equivalents which operate in a similar manner to accomplish a simila r technical purpose. Terms such as "forward", "rea rward", "radia lly", "periphera lly", "upwardly", "downwa rdly", a nd the like are used as words of convenience to provide reference points a nd are notto be construed as limiting terms.

[151] For the purposes of this specification, the term ^'plastic _, shall be construed to mea n a general term for a wide range of synthetic or semisynthetic polymerization products, a nd genera lly consisting of a hydrocarbon-based polymer.

[152] As used herein the term and/or. means a nd. or or., or both.

[153] As used herein ^'(s) . following a noun means the plural a nd/or singula r forms of the noun.

C omprising and Including

[154] In the claims which follow a nd in the preceding description of the invention, except where the context requires otherwise due to express language or necessary implication, the word comprise _, or variations such as comprises _, or comprising_, are used in an inclusive sense, i.e. to specify the presence of the stated features but not to preclude the presence or addition of further features in various embodiments of the invention.

[155] Any one of the terms: including or which includes or that includes as used herein is also an open term that a lso mea ns including at least the elements/features that follow the term, but not excluding others. Thus, including is synonymous with and mea ns comprising.

S cope of Invention

[156] T hus, while there has been described what a re believed to be the preferred embodiments of the invention, those skilled in the art will recognize that other and further modifications may be made thereto without depa rting from the s pirit of the invention, and it is intended to claim a ll such changes and modifications as fa ll within the scope of the invention. For example, any formulas given above a re merely representative of procedures that may be used. Functionality may be added or deleted from the block diagrams and operations may be interchanged among functional blocks. S teps may be added or deleted to methods described within the scope of the present invention.

[157] Although the invention has been described with reference to specific examples, it will be appreciated by those skilled in the art that the invention may be embodied in many other forms.

Industria l Applicability

[158] It is apparent from the above, that the arrangements described are applica ble to the medical diagnosis industries.

Claims

What is c laimed is : 1 ) A system for determining the presence of a movement-related disease in a patient, the system compris ing: a) a processor; b) a memory storage device comprising computer program code a nd being operably connected to the processor; c) an input device operably connected to the processor and adapted to send an input data to the processor when the input device is used by the patient to input data; d) wherein the computer program code is configured for directing the system for: i) receiving patient keystroke feature data; ii) identifying diagnosis data from the patient keystroke feature data, including at least: (1 ) flight time data distribution patterns relating to time periods between keystrokes within the patient keystroke feature data for each ha nd of the patient; and (2) hold time data distribution patterns relating to time periods over which a keystroke is performed within the patient keystroke feature data for each ha nd of the patient; iii) compa ring the identified flight time data distribution patterns and the hold time data distribution patterns for each hand to determine asymmetry data; and iv) compa ring the diagnosis data to simila r control data of healthy persons to determine the likelihood of the presence of Pa rkinson s disease in accordance with the asymmetry data. 2) T he system of claim 1 , wherein a plura lity of machine lea rning models a re applied in para llel to the patient keystroke feature data, either individually or after grouping of related features to determine the likelihood of a presence of the movement- related disease, wherein results of each of the machine learning models a re combined in a weighted meta-classifier. 3) T he system of claim 1 , wherein statistica l attributes of the patient keystroke feature data a re pre-processed a nd dimensionally reduced using at least one of: a) linear discriminant ana lysis; b) kernel extreme learning; a nd c) machine a nd subtractive clustering features weighting; prior to the determination of the likelihood of the presence of the movement- related disease. 4) T he system of claim 1 wherein, the patient keystroke feature data is pre-processed using a combination of: a) linear discriminant ana lysis; b) kernel extreme learning; and c) machine a nd subtractive clustering features weighting; prior to the determination of the likelihood of the presence of the movement- related disease. 5) T he system of claim 1 , wherein the computer program code is further configured to predict the presence of ha nd tremors by executing at least one of the following list of steps: a) comparing the input data with known statistica l input data corresponding to ha nd tremors using a machine learning matching a lgorithm; b) generating a continuous digital data stream from the patient keystroke feature data by removing non-contiguous instances; c) applying pre-processing to the continuous digita l data stream by using a low pass signal filter a nd/or polynomial interpolation to generate a filtered data stream; d) applying Fourier ana lysis to the filtered data stream in order to produce a frequency domain representation and identification of particular frequency components including their res pective power levels indicative of hand tremors; e) creating a series of non-integer sub-sampling decimations from the digita l data stream; applying an ana lysis of variance to the digital data stream in order to identify pa rticula r frequency and/or power components indicative of hand tremors; f) applying a method comprising the steps of: i) receiving contiguous patient keystroke data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke data elements defining feature va lues with associated time stamps; ii) generating at least one or more data arrays from the patient keystroke data elements; iii) for each data array, determine the differences between the feature value of a data element and a n associated va lue of a predetermined wave form at a particular frequency a nd time stamp, over a range of possible tremor frequencies at a plura lity of frequency increments, to generate an a nalysis array; iv) calculating the va riance of the determined differences for each frequency increment; and v) compa ring the varia nce of the particula r frequency to a predetermined thres hold generated from the variances of the frequencies in the frequency ra nge to identify one or more candidate tremor frequencies. 6) A method for determining the presence of Parkins on s disease in a patient, the method compris ing the fol lowing steps : a) receiving patient keystroke feature data; b) identifying diagnosis data from the patient keystroke feature data, including at least: i) flight time data distribution patterns relating to time periods between keystrokes within the patient keystroke feature data for each ha nd of the patient; and ii) hold time data distribution patterns relating to time periods over which a keystroke is performed within the patient keystroke feature data for each hand of the patient; c) comparing the identified flight time data distribution patterns and the hold time data distribution patterns for each hand to determine asymmetry data; a nd d) comparing the diagnosis data to similar control data of healthy persons to determine the likelihood of the presence of Parkinson s disease in accordance with the asymmetry data. 7) T he method of claim 6, wherein the method comprises the step of identifying a plura lity of flight time distribution patterns a nd a plura lity of hold time distribution patterns. 8) T he method of claim 6, wherein the method comprises the step of using the patient keystroke feature data to determine the presence of tremors in a patient. 9) T he method of claim 6, wherein the patient keystroke feature data is pre-processed using at least one of: a) linear discriminant ana lysis; b) kernel extreme learning machine; and c) subtractive clustering features weighting; prior to the determination of the likelihood of the presence of the movement- related disease. 10) T he method of claim 6, wherein the patient keystroke feature data is grouped with machine lea rning applied at least at a n individual or a group level. 1 1 ) T he method of claim 6, wherein a plurality of separate machine learning models is used to create a weighted meta-classifier. 12) T he method of claim 6, wherein a hierarchy of prediction models is used to predict movement-related disease severity. 13) T he method of claim 6, further comprising at least one of the following steps: a) applying machine learning in order to compare the patient keystroke feature data with known input data corresponding to hand tremors; b) creating a continuous digital data strea m from the patient keystroke feature data by removing non-contiguous instances; c) applying pre-processing to the continuous digital data stream by using a low pass signal filter a nd/or polynomial interpolation; d) applying Fourier ana lysis to the continuous digital data stream in order to produce a frequency domain representation and identification of pa rticular frequency components a nd their respective power levels indicative of hand tremors; e) creating a series of non-integer sub-sampling decimations; f) applying a n a nalysis of varia nce to the continuous digital data stream in order to identify particular frequency and/or power components indicative of ha nd tremors; g) a method comprising the steps of: i) receiving contiguous patient keystroke data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke data elements defining feature va lues with associated time stamps; ii) generating at least one or more data arrays from the patient keystroke data elements; iii) for each data array, determine the differences between the feature value of a data element and a n associated va lue of a predetermined wave form at a particular frequency a nd time stamp, over a ra nge of possible tremor frequencies at a plura lity of frequency increments, to generate an a nalysis array; iv) calculating the va riance of the determined differences for each frequency increment; and v) compa ring the varia nce of the particula r frequency to a predetermined threshold generated from the variances of the frequencies in the frequency ra nge to identify one or more candidate tremor frequencies. 14) T he method of claim 13, wherein a plura lity of separate machine learning models is used to create a weighted meta-classifier, to determine the likelihood of the presence of a movement-related disease in the patient in accordance with the presence of at least two separate ca rdinal symptoms. 15) T he method of claim 14, wherein a hiera rchy of prediction models is used to predict the movement-related disease severity. 16) A method for determining the pres ence of hand tremor indicative of a movement" related dis ease in a patient, the method compris ing the steps of: a) receiving contiguous patient keystroke data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke data elements defining feature va lues with associated time stamps; b) generating at least one or more data a rrays from the patient keystroke data elements; c) for each data array, determining the differences between the feature value of a data element a nd an associated va lue of a predetermined wave form at a particular frequency and time stamp, over a range of possible tremor frequencies at a plura lity of frequency increments, to generate a n a nalysis array; d) calculating the varia nce of the determined differences for each frequency increment; and e) comparing the variance of the particular frequency to a predetermined threshold generated from the variances of the frequencies in the frequency range to identify one or more ca ndidate tremor frequencies. 17) T he method of claim 16, wherein the predetermined threshold is a predetermined proportion of the mean of the va riances. 18) T he method of claim 16, wherein the method further comprises the steps of: a) for at least the one or more ca ndidate tremor frequencies and a predetermined range of proximate frequencies, determine a best fit of a predetermined wave form against the feature values for a ra nge of phase angles to identify the phase a ngle of the candidate frequency a nd proximate frequencies; a nd b) comparing the identified phase angles of the at least one or more ca ndidate frequencies with the identified phase angles of the proximate frequencies in order to confirm ca ndidate frequencies as confirmed frequencies. 19) T he method of claim 18, further comprising the step of: a) comparing confirmed frequencies of the data array to the confirmed frequencies of another data a rray. 20) T he method of claim 18, further comprising the steps of: a) determining the sum of the differences between the va riances of the ca ndidate frequencies a nd a predetermined threshold relative to variances at other frequencies; b) determining the number of ca ndidate frequencies; c) diagnosing the presence of a movement related disease based on a combination of the determined sum of the differences and the number of candidate frequencies.21 ) T he method of claim 16, wherein the method comprises replicating the method of any one of claims 16-20 with a new dataset of contiguous patient keystroke data elements, a nd comparing the confirmed frequencies. 22) A system for determining the presence of a movement-related disease in a patient, the system compris ing: a) a processor; b) a memory storage device comprising computer program code and being operably connected to the processor; c) an input device operably connected to the processor and adapted to send an input data to the processor when the input device is used by the patient to input data; d) wherein the computer program code is configured for directing the system for: i) receiving contiguous patient keystroke data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke data elements defining feature va lues with associated time stamps; ii) generating at least one or more data arrays from the patient keystroke data elements; iii) for each data array, determining the differences between the feature value of a data element and a n associated va lue of a predetermined wave form at a particular frequency a nd time stamp, over a ra nge of possible tremor frequencies at a plura lity of frequency increments, to generate an a nalysis array; iv) calculating the va riance of the determined differences for each frequency increment; and v) compa ring the varia nce of the particula r frequency to a predetermined threshold generated from the variances of the frequencies in the frequency ra nge to identify one or more candidate tremor frequencies. 23) T he system of claim 22, wherein the predetermined threshold is a predetermined proportion of the mean of the va riances. 24) T he system of claim 22, wherein the computer progra m code is configured for directing the system for: a) for at least the one or more candidate tremor frequencies and a predetermined range of proximate frequencies, determine a best fit of a predetermined wave form against the feature values for a ra nge of phase angles to identify the phase a ngle of the candidate frequency a nd proximate frequencies; and b) comparing the identified phase angles of the at least one or more ca ndidate frequencies with the identified phase angles of the proximate frequencies in order to confirm ca ndidate frequencies as confirmed frequencies. 25) T he system of claim 22, wherein the computer program code is configured for directing the system for: a) comparing confirmed frequencies of the data array to the confirmed frequencies of another data a rray. 26) T he system of claim 22, wherein the computer progra m code is configured for directing the system for: a) determining the sum of the differences between the variances of the candidate frequencies a nd a predetermined threshold relative to varia nces at other frequencies; b) determining the number of ca ndidate frequencies; c) diagnosing the presence of a movement related disease based on a combination of the determined sum of the differences and the number of candidate frequencies. 27) T he system of claim 22, wherein the computer progra m code is configured for directing the system for: a) replicating the method of any of claims 22-26 with a new dataset of contiguous patient keystroke data elements, a nd compa ring the confirmed frequencies. 28) A method of diagnos ing Parkinson s disease compris ing the steps of: a) receiving contiguous patient keystroke feature data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke feature data elements defining keystroke feature values with associated time stamps; and b) processing the keystroke feature va lues to detect both bradykinesia a nd tremor in a patient. 29) T he method of claim 28, wherein the step of processing the keystroke feature values to detect bradykinesia in a patient comprises the method as claimed in any of claims 6" 16. 30) T he method of claim 28, wherein the step of processing the keystroke feature va lues to detect tremor in a patient comprises the method as claimed in a ny one of claims 17"22. 31 ) A system for determining the presence of a movement-related disease in a patient, the system compris ing: a) a processor; b) a memory storage device comprising computer program code and being operably connected to the processor; c) an input device operably connected to the processor and adapted to send an input data to the processor when the input device is used by the patient to input data; d) wherein the computer program code is configured for directing the system for: i) receiving contiguous patient keystroke feature data elements from at least one or more input devices for at least one hand of a patient, the patient keystroke feature data elements defining keystroke feature values with associated time sta mps; ii) processing the keystroke feature va lues to detect both bradykinesia a nd tremor in a patient. 32) T he system of claim 31 , wherein the computer progra m code is configured for detecting bradykinesia in a patient by directing the system for carrying out steps as claimed in any of claims 6"22. 33) T he system of claim 31 , wherein the computer progra m code is configured for detecting tremor in a patient by directing the system for ca rrying out steps as claimed in any of claims 17 "22. 34) A method for determining the presence of a movement-related disease in a patient, the method compris ing the steps of: a) receiving contiguous patient keystroke data elements from at least one or more input devices; b) generating a plurality of signal a rrays from the patient keystroke data; c) performing a regression analysis on the plurality of signal arrays for a range of possible tremor frequencies at sma ll frequency increments to generate a plurality of signal a na lysis a rrays from each signa l array, each signal ana lysis array including at least one signal a nalysis array element; d) calculating interpolated va lues for each signal a nalysis array element of each signa l analysis array for the particular frequency a nd ela psed time of that signa l analysis array, using at least one of sinusoida l or polynomial interpolation; e) determining the difference between the value of each signal a nalysis array element and the calculated interpolated va lue for that element as a variance; and f) comparing the variance of each signa l ana lysis a rray element at a particular ta rget tremor frequency with the variance of signal analysis array elements at other ta rget tremor frequencies to detect one or more selected from: i) one or more domina nt tremor frequencies; a nd ii) one or more domina nt tremor frequency ranges; that are indicative of hand tremor in the patient. 35) A system for determining the presence of a movement-related disease in a patient, the system compris ing: a) a processor; b) a memory storage device comprising computer program code and being operably connected to the processor; c) an input device operably connected to the processor and adapted to send an input data to the processor when the input device is used by the patient to input data; d) wherein the computer program code is configured for directing the system for: i) receiving contiguous patient keystroke data elements from at least one or more input devices; ii) generating a plura lity of signal a rrays from the patient keystroke data; iii) performing a regression ana lysis on the plura lity of signal arrays for a range of possible tremor frequencies at small frequency increments to generate a plurality of signa l ana lysis a rrays from each signa l array, each signa l ana lysis a rray including at least one signal a nalysis array element; iv) calculating interpolated values for each signal analysis a rray element of each signal analysis array for the particula r frequency and elapsed time of that signal ana lysis a rray, using at least one of sinusoidal or polynomial interpolation; v) determining the difference between the va lue of each signal analysis array element and the ca lculated interpolated value for that element as a va riance; a nd vi) compa ring the varia nce of each signal a nalysis a rray element at a particular ta rget tremor frequency with the va riance of signal analysis array elements at other target tremor frequencies to detect one or more selected from:

(1 ) one or more dominant tremor frequencies; and

(2) one or more domina nt tremor frequency ra nges; that a re indicative of ha nd tremor in the patient.

36) A method for determining the presence of hand tremor indicative of a movement^" related dis ease in a patient, the method compris ing the steps of:

a) receiving contiguous patient keystroke data elements from at least one or more input devices for at least one ha nd of a patient, the patient keystroke data elements defining feature va lues with associated time stamps;

b) generating at least one or more data a rrays from the patient keystroke data

elements; and

c) applying regression a nalysis of a sinusoida l wave to each data array for a range of incrementa l frequencies, over a range of phase angles for each frequency, to establish which of the frequency increments provides a best fit to the data array to establish one or more dominant frequencies.