CN116189898A

CN116189898A - Method and device for training predictive model for predicting parkinsonism risk level

Info

Publication number: CN116189898A
Application number: CN202310455842.5A
Authority: CN
Inventors: 王振常; 魏璇; 赵鹏飞; 黄继随; 魏巍
Original assignee: Beijing Friendship Hospital
Current assignee: Beijing Friendship Hospital
Priority date: 2023-04-25
Filing date: 2023-04-25
Publication date: 2023-05-30
Anticipated expiration: 2043-04-25
Also published as: CN116189898B

Abstract

The application discloses a method and a device for training a prediction model for predicting the risk level of parkinsonism. The method comprises the following steps: acquiring diffusion kurtosis imaging reflecting brain interval differences of a subject, and extracting target parameter images related to Parkinson's disease from the diffusion kurtosis imaging; performing random assignment operation on the target parameter image to obtain a random parameter image; based on the target parameter image and the random parameter image, performing feature extraction by using a first sub-model to construct a positive example sample and a negative example sample; calculating a first loss function of the prediction model according to the positive example sample and the negative example sample; performing a classification operation using the second sub-model and calculating a second loss function of the predictive model based at least on the target parametric image; and training the predictive model using the first loss function and the second loss function to train the predictive model for predicting the risk level of parkinson's disease. By utilizing the scheme of the method and the device, the accuracy of the prediction result can be improved.

Description

Method and device for training predictive model for predicting parkinsonism risk level

Technical Field

The present application relates generally to the field of artificial intelligence technology. More particularly, the present application relates to a method and apparatus for training a predictive model for predicting the risk level of parkinson's disease. Further, the present application relates to an apparatus for predicting a risk level for parkinson's disease and a computer readable storage medium.

Background

Parkinson's Disease ("PD") is a common central nervous system degenerative Disease of middle-aged and elderly people, and its clinical manifestations mainly include resting tremor, bradykinesia, myotonia and postural gait disorder, with symptoms such as cognitive dysfunction. Studies have shown that patients with Parkinson's disease also have altered brain function connections to their brains.

Currently, there is a method for predicting the risk of parkinson's disease by acquiring brain images of parkinson's disease patients and performing feature extraction by inputting the brain images into a neural network model. However, the existing method relies on convolutional neural networks alone, so that it is difficult to extract the distinguishing characteristics in the brain function connection of parkinsonism patients, and the prediction result is inaccurate. In addition, when training the convolutional neural network, the conventional mode is to perform supervision training first and then train the convolutional neural network, so that the loss of the front training stage and the back training stage cannot be considered, and the accuracy of the trained convolutional neural network is not high.

In view of the foregoing, it is desirable to provide a solution for training a prediction model for predicting the risk level of parkinson's disease, so as to improve the accuracy of the prediction model, thereby sufficiently extracting the discriminatory features in the brain function connection of parkinson's disease patients and improving the accuracy of the prediction result.

Disclosure of Invention

In order to solve at least one or more of the technical problems mentioned above, the present application proposes, in various aspects, a solution for training a predictive model for predicting the risk level of parkinson's disease.

In a first aspect, the present application provides a method for training a predictive model for predicting a level of risk of parkinson's disease, wherein the predictive model comprises a first sub-model and a second sub-model, and the method comprises: acquiring diffusion kurtosis imaging reflecting brain interval differences of a subject, and extracting a target parameter image related to parkinson's disease from the diffusion kurtosis imaging; performing random assignment operation on the target parameter image to obtain a random parameter image; based on the target parameter image and the random parameter image, performing feature extraction by using the first sub-model to construct a positive example sample and a negative example sample; calculating a first loss function of the prediction model according to the positive example sample and the negative example sample; performing a classification operation using the second sub-model and calculating a second loss function of the predictive model based at least on the target parametric image; and training the predictive model using the first and second loss functions to train a predictive model that predicts a level of risk of parkinson's disease.

In one embodiment, wherein the target parameter image and the random parameter image comprise at least two types of images of an average kurtosis image, an axial kurtosis image, and a radial kurtosis image.

In another embodiment, wherein performing feature extraction using the first sub-model to construct a positive example sample and a negative example sample based on the target parametric image and the random parametric image comprises: the first sub-model is used for extracting the characteristics of the target parameter image and the random parameter image respectively so as to obtain first characteristic quantity and second characteristic quantity corresponding to various target parameter images and various random parameter images; and constructing the positive example sample and the negative example sample based on the first characteristic quantity and the second characteristic quantity corresponding to the various target parameter images and the various random parameter images.

In yet another embodiment, wherein extracting the positive example sample and the negative example sample based on the first feature amount and the second feature amount corresponding to the respective types of target parameter images and the respective types of random parameter images includes: forming a set by the first characteristic quantity corresponding to the target parameter image and the second characteristic quantity corresponding to the same type of image in the random parameter image so as to construct the positive sample; and combining the first characteristic quantity corresponding to the target parameter image with the second characteristic quantity corresponding to other types of images in the random parameter image to construct the counterexample sample.

In yet another embodiment, wherein calculating the first loss function of the predictive model from the positive example sample and the negative example sample comprises: calculating similarity based on the positive example sample and the negative example sample; and calculating a first loss function of the prediction model according to the similarity result.

In yet another embodiment, the method further comprises: acquiring cognitive scale information related to the Parkinson's disease; and performing a classification operation using the second sub-model and calculating a second loss function of the predictive model based on the target parameter image and the cognitive scale information.

In yet another embodiment, wherein training the predictive model with the first and second loss functions to train a predictive model that predicts a level of risk of parkinson's disease comprises: performing a weighted sum operation on the first and second loss functions to obtain a total loss function; and training the prediction model by using the total loss function to train a prediction model for predicting the risk level of the Parkinson's disease.

In a second aspect, the present application provides an apparatus for predicting a level of risk of parkinson's disease, comprising: a processor; a memory storing program instructions for predicting a level of risk of parkinson's disease, which when executed by the processor, cause the apparatus to: acquiring diffusion kurtosis imaging reflecting brain interval differences of a subject, and extracting a target parameter image related to parkinson's disease from the diffusion kurtosis imaging; and inputting the target parameter image into a prediction model trained according to the plurality of embodiments in the first aspect to predict, so as to output a prediction result of predicting the risk level of the parkinsonism.

In a third aspect, the present application provides an apparatus for training a predictive model for predicting a level of risk of parkinson's disease, comprising: a processor; a memory storing program instructions for training a predictive model for predicting a risk level of parkinson's disease, which program instructions, when executed by the processor, cause the apparatus to carry out the plurality of embodiments of the aforementioned first aspect.

In a fourth aspect, the present application provides a computer readable storage medium having stored thereon computer readable instructions for predicting a level of risk of parkinson's disease and for training a prediction model of predicting a level of risk of parkinson's disease, which when executed by one or more processors, implement the operations performed by the apparatus of the first aspect, embodiments of the second aspect, and the like.

By the scheme provided above for training the prediction model for predicting the risk level of parkinson's disease, the embodiment of the application constructs a positive example sample and a negative example sample by extracting a target parameter image related to parkinson's disease from diffusion kurtosis imaging and performing a random assignment operation on the target parameter image to obtain a random parameter image, and then performing feature extraction on the target parameter image and the random parameter image using the first sub-model. And calculating a first loss function of the prediction model through the positive example sample and the negative example sample, calculating a second loss function based on the target parameter image by combining the second sub-model, and training the prediction model by using the first loss function and the second loss function. That is, in the embodiment of the application, the positive and negative samples are constructed based on the target parameter image and the random parameter image to perform contrast supervision learning and obtain the first loss, and the prediction model is trained by combining the second loss, so that the first loss and the second loss are simultaneously considered, the precision of the prediction model is improved, the identification information related to the parkinsonism is fully extracted, and the accuracy of the prediction model is improved.

Further, the embodiment of the application further obtains the cognition scale information related to the parkinsonism to perform the second loss function based on the target parameter image and the cognition scale information, namely combining the two types of data, so as to further provide the accuracy of the prediction model. Based on the above, the prediction model trained by the scheme of the embodiment of the application is used for predicting the risk level of the Parkinson's disease, so that an accurate prediction result can be obtained.

Drawings

The above, as well as additional purposes, features, and advantages of exemplary embodiments of the present application will become readily apparent from the following detailed description when read in conjunction with the accompanying drawings. Several embodiments of the present application are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which like reference numerals refer to similar or corresponding parts and in which:

FIG. 1 is an exemplary flow chart illustrating a method for training a predictive model for predicting a risk level of Parkinson's disease in accordance with an embodiment of the present application;

FIG. 2 is an exemplary diagram illustrating training of a predictive model for predicting a risk level of Parkinson's disease in accordance with an embodiment of the present application;

FIG. 3 is a further exemplary diagram illustrating training of a predictive model for predicting a risk level of Parkinson's disease in accordance with an embodiment of the present application;

FIG. 4 is an exemplary block diagram illustrating an apparatus for predicting a level of risk of Parkinson's disease in accordance with an embodiment of the present application; and

fig. 5 is an exemplary block diagram illustrating an apparatus for training a predictive model for predicting a risk level of parkinson's disease according to an embodiment of the application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the embodiments described in this specification are only some embodiments provided herein for the purpose of facilitating a clear understanding of the solution and meeting legal requirements, and not all embodiments of the application may be implemented. All other embodiments, which can be made by those skilled in the art without the exercise of inventive faculty, are intended to be within the scope of the present application, based on the embodiments disclosed in this specification.

Fig. 1 is an exemplary flow chart illustrating a method 100 for training a predictive model for predicting a level of risk of parkinson's disease according to an embodiment of the application. Wherein the predictive model may include a first sub-model and a second sub-model. As shown in fig. 1, at step 101, diffusion kurtosis imaging reflecting a difference between brain regions of a subject is acquired, and a target parameter image related to parkinson's disease is extracted from the diffusion kurtosis imaging. As known to those skilled in the art, diffusion kurtosis imaging (Diffusion Kurtosis Imaging, "DKI") is a magnetic resonance imaging (Magnetic Resonance Imaging, "MRI") technique that reflects differences between brain regions, and is used to quantitatively analyze brain structures of PD patients, extract target parameter images related to parkinson's disease in diffusion kurtosis imaging, and reflect feature information of functional connection of the brain, so as to predict risk level of parkinson's disease.

In some embodiments, the target parameter image may include, but is not limited to, at least two of an average Kurtosis (MK) image, an Axial Kurtosis (AK) image, and a Radial Kurtosis (RK) image. For example, the target parameter image may be, for example, an average Dispersion value (Mean Dispersion, "MD"), a partial anisotropy value (Fractional Anisotropy, "FA"), or the like, and the present application is not limited in this respect. Preferably, the target image parameters of the present application may include MK image, AK image, and RK image.

Based on the target parameter image obtained as described above, at step 102, a random assignment operation is performed on the target parameter image to obtain a random parameter image. In one implementation scenario, the random parametric image may be obtained by randomly assigning gray values of one or more pixels in the target parametric image (e.g., MK image, AK image, and RK image) to 0 to perform a random assignment operation on the target parametric image. The random parametric image thus obtained also correspondingly comprises at least two of MK image, AK image and RK image.

Next, at step 103, feature extraction is performed using the first sub-model based on the target parameter image and the random parameter image to construct a positive example sample and a negative example sample. In one embodiment, first, a first sub-model may be used to perform feature extraction on the target parameter image and the random parameter image respectively, so as to obtain first feature amounts and second feature amounts corresponding to each type of target parameter image and each type of random parameter image; and then constructing a positive example sample and a negative example sample based on the first characteristic quantity and the second characteristic quantity corresponding to the various target parameter images and the various random parameter images. In one implementation scenario, the aforementioned first sub-model may be, for example, a convolutional neural network. Taking the target parameter image and the random parameter image as MK image, AK image and RK image as examples, under the scene, respectively inputting the target parameter image and the random parameter image into a convolutional neural network for feature extraction, and obtaining first feature quantities corresponding to the MK image, AK image and RK image before random assignment operation and second feature quantities corresponding to the MK image, AK image and RK image after random assignment operation.

In some embodiments, the target parametric image and the random parametric image may be pre-processed to process each type of image into an image (MK image, AK imageImage and RK image) of the same size. Taking MK images as an example, it is understood that MK markers exist in 246 brain regions divided based on the human brain network group (The Human Brainnetome), for each brain region, the origin moment and center moment of the first 500 th order of MK distribution can be calculated. Wherein, the first-order central moment of any one distribution is 0, and the 0-order arbitrary moment is equal, namely, the actual meaningful data is only 1000, thus MK images can be formed

Is a matrix of (a) in the matrix. In addition, MK markers are also present in 150 brain regions divided based on anatomical automatic calibration (Anatomical Automatic Labeling, "AAL") templates, whereby MK images can also be formed +.>

Is a matrix of (a) in the matrix. By combining the two dividing modes, MK image can be processed into +.>

Is a matrix of (a) in the matrix. Similar to the MK image described above, AK and RK images can be processed to +.>

Whereby the pretreated MK image, AK image and RK image form +.>

Three channels of data.

After the first characteristic quantity and the second characteristic quantity corresponding to various target parameter images and various random parameter images are obtained, a positive example sample and a negative example sample can be constructed. Specifically, a set is formed by the first characteristic quantity corresponding to the target parameter image and the second characteristic quantity corresponding to the same type of image in the random parameter image so as to construct a positive example sample, and a set is formed by the first characteristic quantity corresponding to the target parameter image and the second characteristic quantity corresponding to other types of images in the random parameter image so as to construct a negative example sample. That is, in the embodiment of the present application, the first feature quantity and the second feature quantity corresponding to the same type of image of the target parameter image (i.e., before the random assignment operation) and the random parameter image (i.e., after the random assignment operation) are taken as positive examples, and the first feature quantity corresponding to the target parameter image (i.e., before the random assignment operation) and the second feature quantity corresponding to the different type of image of the random parameter image (i.e., after the random assignment operation) are taken as negative examples.

For example, in one exemplary scenario, it is assumed that first feature amounts corresponding to MK image, AK image, and RK image in the target parameter image are respectively noted as

、/>

And->

Second feature quantities corresponding to MK image, AK image and RK image in the random parameter image are respectively marked as +.>

、/>

And->

. For MK image, first feature quantity +.>

Second feature quantity corresponding to MK image in random parameter image +.>

As a positive example sample; first feature quantity +_corresponding to MK image in target parameter image>

And a second feature quantity corresponding to the AK image and the RK image in the random parameter image

、/>

As a counterexample sample. />

Similarly, a first feature quantity corresponding to an AK image in the target parameter image

Second feature quantity corresponding to AK image in random parameter image +.>

As a positive example sample; first feature quantity corresponding to AK image in target parameter image +.>

Second feature quantity +.2 corresponding to MK image and RK image in random parameter image>

、/>

As a counterexample sample. First feature quantity corresponding to RK image in target parameter image>

Second feature quantity corresponding to RK image in random parameter image +.>

As a positive example sample; first feature quantity corresponding to RK image in target parameter image>

Second feature quantity +.2 corresponding to MK image and AK image in random parameter image>

、/>

As a counterexample sample.

In constructing the above positive example sample and the negative exampleAfter the sample, at step 104, a first loss function of the predictive model is calculated from the positive example sample and the negative example sample. In one embodiment, similarity is first calculated based on the positive example sample and the negative example sample to calculate a first loss function of the predictive model based on the similarity results. Specifically, the similarity between positive examples can be calculated first, e.g., as

The method comprises the steps of carrying out a first treatment on the surface of the Then the similarity between counterexample samples is calculated, e.g. noted assim(/>

To calculate a first loss function of the predictive model based on the foregoing similarity results. In one implementation scenario, the first loss function of the predictive model may be calculated by the following formula ∈>

：

（1）

Wherein, the liquid crystal display device comprises a liquid crystal display device,

representing first characteristic quantity corresponding to various images in target parameter image,/for>

Representing second characteristic quantities corresponding to various images in the random parameter image,i=1, 2, …, N, and +.>

. I.e.)>

Representing a positive sample, ++>

Representing a counterexample sample, ++>

Representing the similarity between the samples of the positive examples,sim(/>

representing the similarity between counterexample samples, and +.>

，sim(/>

. In addition, a->

Expressed as natural constant->

An exponential function of the base +.>

For indicating functions, i.e. if->

If true, the value is 1; otherwise, the value is 0.

Further, at step 105, based at least on the target parametric image, a classification operation is performed using the second sub-model and a second loss function of the predictive model is calculated. In one embodiment, the second sub-model may be, for example, a fully connected network, and may thereafter contain a softmax function to output the prediction. In one implementation, the aforementioned second loss function

Can be expressed as follows:

（2）

sample number representing target parameter image, +.>

Representing the number of category labels->

And->

Respectively representing the labels of the samples and the corresponding predicted values. In the embodiment of the present application, the aforementioned category label may be a classification level of the parkinson's disease risk level, such as low, medium or high.

According to the first and second loss functions obtained as described above, at step 106, a predictive model is trained using the first and second loss functions to train a predictive model that predicts the risk level of parkinson's disease. In one embodiment, a weighted sum operation may be performed on the first and second loss functions to obtain a total loss function, which is used to train a predictive model to predict the risk level of parkinson's disease. For example, assume that the total loss function is recorded as

Then the total loss function->

Can be expressed as:

（3）

representing a first loss function, ">

Representing a second loss function, ">

Representing the weight coefficient. Based on the obtained total loss function, training of a prediction model for predicting the risk level of the Parkinson disease can be realized。/>

As can be seen from the above description, in the embodiment of the present application, the first feature quantity and the second feature quantity corresponding to each type of image are obtained by performing feature extraction on the target parameter image and the random parameter image, and the first feature quantity and the second feature quantity corresponding to the same type of image in the target parameter image and the random parameter image are taken as positive examples, and the first feature quantity corresponding to each type of image in the target parameter image and the second feature quantity corresponding to other types of image in the random parameter image are taken as negative examples. Then, contrast learning is performed based on the positive example sample and the negative example sample and a first loss function is constructed to make full use of the intrinsic information of the data, so that the discriminative information related to parkinson's disease can be extracted effectively. Further, a total loss function is obtained through combining a second loss function calculated by a second sub-model based on the target parameter image, so that a prediction model for predicting the risk level of the Parkinson's disease is trained, and the first loss and the second loss are simultaneously considered. Based on the method, the accuracy of the prediction model can be improved, so that when the trained prediction model is used for predicting the risk level of the Parkinson's disease, an accurate prediction result can be obtained.

In one embodiment, the method further includes obtaining cognitive scale information associated with parkinson's disease, and based on the target parameter image and the cognitive scale information, performing a classification operation using a second sub-model and calculating a second loss function of the predictive model. In some embodiments, the aforementioned cognitive scale information may include at least Becky anxiety scale information and/or Becky depression scale information, and the aforementioned cognitive scale information may also include a simple mental state examination scale (Mini-mental State Examination, "MMSE") or the like. In addition, gait parameters (such as parameters of pace speed, stride length or stride rhythm) can be obtained, and a second loss function of the prediction model is calculated in combination with the gait parameters. Specifically, normalization operation is performed on the cognitive scale information, and the data after normalization operation is fused with the first feature quantity after feature extraction of the target parameter image. And then, inputting the fused data into a second sub-model for classification operation, and calculating a second loss function of the prediction model.

Fig. 2 is an exemplary schematic diagram illustrating training of a predictive model for predicting a risk level of parkinson's disease according to an embodiment of the application. As shown in fig. 2, the extracted target parameter image 201 and the random parameter image 202 obtained after the random assignment operation (for example, indicated by an arrow a) are respectively input into the first sub-model 203 for feature extraction. In one exemplary scenario, the aforementioned object parameter image 201 comprises MK image 201-1, AK image 201-2, and RK image 201-3, and the aforementioned random parameter image 202 comprises MK image 202-1, AK image 202-2, and RK image 202-3 after a random assignment operation. In this scenario, after feature extraction via the first sub-model 203, MK image 201-1, AK image 201-2 and RK image 201-3 in the target parameter image 201 respectively correspond to obtain first feature amounts

First characteristic quantity->

And a first characteristic quantity->

MK image 202-1, AK image 202-2 and RK image 202-3 after random assignment operation respectively obtain second feature quantity +.>

Second characteristic quantity->

And second characteristic quantity->

. In one embodiment, the aforementioned first sub-model 203 may be, for example, a convolutional neural network.

Further, taking MK image as an example, a first feature quantity corresponding to MK image 201-1 in target parameter image 201 is obtained

And after the random assignment operationSecond feature quantity +.corresponding to MK image 202-1>

As a Positive example sample (i.e., "Positive"). In addition, a first feature quantity corresponding to MK image 201-1 in target parameter image 201 is to be +.>

Second feature quantity +.corresponding to each of AK image 202-2 and RK image 202-3 after random assignment operation>

And second characteristic quantity->

As a counterexample sample (i.e., "Negative"). Based on the constructed positive and negative examples, a first loss function of the predictive model can be predicted according to the above formula (1)>

。

As further shown in the figure, the embodiment of the present application also obtains the first feature quantities by respectively corresponding to MK image 201-1, AK image 201-2 and RK image 201-3 in the target parameter image 201

First characteristic quantity->

And a first characteristic quantity->

Input to the second sub-model 204 for classification operations. In one embodiment, the aforementioned second sub-model 204 may be, for example, a fully connected network, and may thereafter include a softmax function to output the prediction. Wherein during the aforesaid classification operation, the second loss function of the predictive model can be calculated by the above formula (2)>

. According to calculationsFirst loss function->

And a second loss function->

The total loss function +.can be obtained based on the above equation (3)>

To use the total loss function +.>

And training a prediction model for predicting the risk level of the parkinsonism.

Fig. 3 is a further exemplary schematic diagram illustrating training of a predictive model for predicting a risk level of parkinson's disease according to an embodiment of the application. As shown in fig. 3, the extracted target parameter image 201 and the random parameter image 202 obtained after the random assignment operation (for example, indicated by an arrow a) are respectively input into the first sub-model 203 for feature extraction. In one exemplary scenario, the aforementioned object parameter image 201 comprises MK image 201-1, AK image 201-2, and RK image 201-3, and the aforementioned random parameter image 202 comprises MK image 202-1, AK image 202-2, and RK image 202-3 after a random assignment operation. In this scenario, after feature extraction via the first sub-model 203, MK image 201-1, AK image 201-2 and RK image 201-3 in the target parameter image 201 respectively correspond to obtain first feature amounts

First characteristic quantity->

And a first characteristic quantity->

Second characteristic quantity/>

And second characteristic quantity->

Second feature quantity ++corresponding to MK image 202-1 after random assignment operation>

And second characteristic quantity->

。

As further shown, the embodiment of the present application also obtains the cognitive scale information 205. In this case, the present embodiment obtains the first feature quantity by respectively corresponding the MK image 201-1, the AK image 201-2, and the RK image 201-3 in the target parameter image 201

First characteristic quantity/>

And a first characteristic quantity->

And the cognition scale information 205 after the normalization operation is performed is input to the second sub-model 204 for classification operation. In one embodiment, the aforementioned second sub-model 204 may be, for example, a fully connected network, and may thereafter include a softmax function to output the prediction. Wherein during the aforesaid classification operation, the second loss function of the predictive model can be calculated by the above formula (2)>

. From the calculated first loss function

And a second loss function->

The total loss function +.can be obtained based on the above equation (3)>

To use the total loss function +.>

Fig. 4 is an exemplary block diagram illustrating an apparatus 400 for predicting a level of risk of parkinson's disease according to an embodiment of the present application. As shown in fig. 4, the apparatus 400 may include a processor 401 and a memory 402. The processor 401 may include, for example, a general-purpose processor ("CPU") or a special-purpose graphics processor ("GPU"), and the memory 402 stores therein program instructions executable on the processor. In some embodiments, the aforementioned Memory 402 may include, but is not limited to, a resistive Random Access Memory RRAM (Resistive Random Access Memory), a dynamic Random Access Memory DRAM (Dynamic Random Access Memory), a Static Random Access Memory SRAM (Static Random-Access Memory), and an enhanced dynamic Random Access Memory EDRAM (Enhanced Dynamic Random Access Memory).

Further, the memory 402 may store program instructions for predicting a level of risk of parkinson's disease, which when executed by the processor 401, cause the apparatus 400 to: diffusion kurtosis imaging reflecting brain interval differences of a subject is acquired, and a target parameter image related to parkinson's disease is extracted from the diffusion kurtosis imaging to be input to a prediction model trained according to an embodiment of the present application for prediction, so as to output a prediction result of predicting a risk level of parkinson's disease. In one implementation scenario, the prediction model may output probabilities corresponding to low, medium and high levels of risk of parkinson's disease, and determine a class with the highest probability value as a prediction result for predicting the risk level of parkinson's disease.

Fig. 5 is an exemplary block diagram illustrating an apparatus 500 for training a predictive model for predicting a level of risk of parkinson's disease according to an embodiment of the application.

As shown in fig. 5, the apparatus of the present application may include a central processing unit or central processing unit ("CPU") 511, which may be a general purpose CPU, a special purpose CPU, or other information processing and program running execution unit. Further, device 500 may also include a mass memory 512 and a read only memory ("ROM") 513, where mass memory 512 may be configured to store various types of data including various and diffuse kurtosis imaging, target parameter images, random parameter images, algorithm data, intermediate results, and various programs needed to operate device 500. ROM 513 may be configured to store data and instructions necessary to power-on self-test for device 500, initialization of functional modules in the system, drivers for basic input/output of the system, and boot the operating system.

Optionally, the device 500 may also include other hardware platforms or components, such as a tensor processing unit ("TPU") 514, a graphics processing unit ("GPU") 515, a field programmable gate array ("FPGA") 516, and a machine learning unit ("MLU") 517, as shown. It will be appreciated that while various hardware platforms or components are shown in device 500, this is by way of example only and not limitation, and that one of skill in the art may add or remove corresponding hardware as desired. For example, device 500 may include only a CPU, associated memory devices, and interface devices to implement the methods of the present application for training a predictive model for predicting a level of risk of parkinson's disease.

In some embodiments, to facilitate the transfer and interaction of data with external networks, the device 500 of the present application further comprises a communication interface 518, such that it may be connected to a local area network/wireless local area network ("LAN/WLAN") 505 via the communication interface 518, and further to a local server 506 or to the Internet ("Internet") 507 via the LAN/WLAN. Alternatively or additionally, the device 500 of the present application may also be directly connected to the internet or cellular network via the communication interface 518 based on wireless communication technology, such as wireless communication technology based on generation 3 ("3G"), generation 4 ("4G"), or generation 5 ("5G"). In some application scenarios, the device 500 of the present application may also access the server 508 and database 509 of the external network as needed to obtain various known algorithms, data, and modules, and may store various data remotely, such as various types of data or instructions for presenting, for example, diffuse kurtosis imaging, target parametric images, random parametric images, etc.

The peripheral devices of the apparatus 500 may include a display device 502, an input device 503, and a data transmission interface 504. In one embodiment, display device 502 may, for example, include one or more speakers and/or one or more visual displays configured for voice prompts and/or visual image display of a predictive model of the present application that is trained to predict the risk level of parkinson's disease. The input device 503 may include other input buttons or controls, such as a keyboard, mouse, microphone, gesture-capturing camera, etc., configured to receive input of audio data and/or user instructions. The data transfer interface 504 may include, for example, a serial interface, a parallel interface, or a universal serial bus interface ("USB"), a small computer system interface ("SCSI"), serial ATA, fireWire ("FireWire"), PCI Express, and high definition multimedia interface ("HDMI"), etc., configured for data transfer and interaction with other devices or systems. According to aspects of the subject application, the data transmission interface 504 may receive diffusion kurtosis imaging from, for example, a magnetic resonance acquisition and transmit data or results including diffusion kurtosis imaging or various other types to the device 500.

The above-described CPU 511, mass memory 512, ROM 513, TPU 514, GPU 515, FPGA 516, MLU 517, and communication interface 518 of the device 500 of the present application may be interconnected by a bus 519 and data interaction with peripheral devices may be accomplished by the bus. In one embodiment, CPU 511 may control other hardware components in device 500 and its peripherals through this bus 519.

An apparatus for training a predictive model for predicting the risk level of parkinson's disease that may be used to perform the present application is described above in connection with fig. 5. It is to be understood that the device structure or architecture herein is merely exemplary, and that the implementation and implementation entities of the present application are not limited thereto, but that changes may be made without departing from the spirit of the present application.

Those skilled in the art will also appreciate from the foregoing description, taken in conjunction with the accompanying drawings, that embodiments of the present application may also be implemented in software programs. The present application thus also provides a computer readable storage medium. The computer readable storage medium may be used to implement the method for training the predictive model for predicting the risk level of parkinson's disease described in connection with fig. 1 of the present application and the operations performed by the apparatus for predicting the risk level of parkinson's disease described in connection with fig. 4.

It should be noted that although the operations of the methods of the present application are depicted in the drawings in a particular order, this does not require or imply that the operations must be performed in that particular order or that all of the illustrated operations be performed in order to achieve desirable results. Rather, the steps depicted in the flowcharts may change the order of execution. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step to perform, and/or one step decomposed into multiple steps to perform.

It should be understood that when the terms "first," "second," "third," and "fourth," etc. are used in the claims, the specification and the drawings of this application, they are used merely to distinguish between different objects and not to describe a particular sequence. The terms "comprises" and "comprising," when used in the specification and claims of this application, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

It is also to be understood that the terminology used in the description of the present application is for the purpose of describing particular embodiments only, and is not intended to be limiting of the application. As used in the specification and claims of this application, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should be further understood that the term "and/or" as used in the present specification and claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.

Although the embodiments of the present application are described above, the content is only an example adopted for understanding the present application, and is not intended to limit the scope and application scenario of the present application. Any person skilled in the art can make any modifications and variations in form and detail without departing from the spirit and scope of the disclosure, but the scope of the disclosure is still subject to the scope of the claims.

Claims

1. A method for training a predictive model for predicting a level of risk of parkinson's disease, wherein the predictive model comprises a first sub-model and a second sub-model, and the method comprises:

acquiring diffusion kurtosis imaging reflecting brain interval differences of a subject, and extracting a target parameter image related to parkinson's disease from the diffusion kurtosis imaging;

performing random assignment operation on the target parameter image to obtain a random parameter image;

based on the target parameter image and the random parameter image, performing feature extraction by using the first sub-model to construct a positive example sample and a negative example sample;

calculating a first loss function of the prediction model according to the positive example sample and the negative example sample;

performing a classification operation using the second sub-model and calculating a second loss function of the predictive model based at least on the target parametric image; and

and training the prediction model by using the first loss function and the second loss function so as to train the prediction model for predicting the risk level of the parkinsonism.

2. The method of claim 1, wherein the target parameter image and the random parameter image comprise at least two types of images of an average kurtosis image, an axial kurtosis image, and a radial kurtosis image.

3. The method of claim 2, wherein performing feature extraction using the first sub-model based on the target parametric image and the random parametric image to construct a positive example sample and a negative example sample comprises:

the first sub-model is used for extracting the characteristics of the target parameter image and the random parameter image respectively so as to obtain first characteristic quantity and second characteristic quantity corresponding to various target parameter images and various random parameter images; and

and constructing the positive example sample and the negative example sample based on the first characteristic quantity and the second characteristic quantity corresponding to the various target parameter images and the various random parameter images.

4. The method according to claim 3, wherein extracting the positive example sample and the negative example sample based on the first feature amount and the second feature amount corresponding to the respective types of target parameter images and the respective types of random parameter images includes:

forming a set by the first characteristic quantity corresponding to the target parameter image and the second characteristic quantity corresponding to the same type of image in the random parameter image so as to construct the positive sample; and

and combining the first characteristic quantity corresponding to the target parameter image with the second characteristic quantity corresponding to other types of images in the random parameter image to form a set so as to construct the counterexample sample.

5. The method of claim 4, wherein calculating a first loss function of the predictive model from the positive example sample and the negative example sample comprises:

calculating similarity based on the positive example sample and the negative example sample; and

a first loss function of the predictive model is calculated based on the similarity results.

6. The method as recited in claim 2, further comprising:

acquiring cognitive scale information related to the Parkinson's disease; and

based on the target parametric image and the cognitive scale information, performing a classification operation using the second sub-model and calculating a second loss function of the predictive model.

7. The method of claim 2, wherein training the predictive model with the first and second loss functions to train a predictive model that predicts a level of risk of parkinson's disease comprises:

performing a weighted sum operation on the first and second loss functions to obtain a total loss function; and

and training the prediction model by utilizing the total loss function so as to train the prediction model for predicting the risk level of the parkinsonism.

8. An apparatus for predicting a level of risk of parkinson's disease, comprising:

a processor;

a memory storing program instructions for predicting a level of risk of parkinson's disease, which when executed by the processor, cause the apparatus to:

acquiring diffusion kurtosis imaging reflecting brain interval differences of a subject, and extracting a target parameter image related to parkinson's disease from the diffusion kurtosis imaging; and

inputting the target parameter image into a prediction model trained according to the method of any one of claims 1-7 for prediction, so as to output a prediction result of predicting the risk level of the parkinsonism.

9. An apparatus for training a predictive model for predicting the risk level of parkinson's disease, comprising:

a processor;

a memory storing program instructions for training a predictive model for predicting a risk level of parkinson's disease, which program instructions, when executed by the processor, cause the apparatus to carry out the method according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon computer readable instructions for predicting a level of risk of parkinson's disease and for training a prediction model for predicting a level of risk of parkinson's disease, which computer readable instructions, when executed by one or more processors, implement the operations performed by the apparatus of claim 8 and the method of any of claims 1-7.