TWI469764B - System, method, recording medium and computer program product for calculating physiological index - Google Patents
System, method, recording medium and computer program product for calculating physiological index Download PDFInfo
- Publication number
- TWI469764B TWI469764B TW100146537A TW100146537A TWI469764B TW I469764 B TWI469764 B TW I469764B TW 100146537 A TW100146537 A TW 100146537A TW 100146537 A TW100146537 A TW 100146537A TW I469764 B TWI469764 B TW I469764B
- Authority
- TW
- Taiwan
- Prior art keywords
- data
- metadata
- windows
- calculating
- physiological parameter
- Prior art date
Links
Landscapes
- Medical Treatment And Welfare Office Work (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Description
本揭露是有關於一種運算生理參數指標的系統、方法、記錄媒體及電腦程式產品。The disclosure relates to a system, method, recording medium and computer program product for calculating physiological parameter indicators.
利用資訊技術，蒐集各種生理訊號並且分析個案生理機能健康狀態，是醫療與資訊領域所共同合作研究的研究課題，例如心電圖(electrocardiogram,ECG)訊號的分析由於其能直接反應心臟機能狀態，在心血管相關疾病分析中一直是相當重要的探索議題。The use of information technology to collect various physiological signals and analyze the health status of individual cases is a research topic in the field of medical and information research. For example, the analysis of electrocardiogram (ECG) signals is directly in response to cardiac function status in cardiovascular disease. Related disease analysis has always been a very important topic of exploration.
疾病之癥兆大多會在生理器官運作與律動的變異性上呈現些微的差異，而針對生理訊號的監控與分析雖然已有許多國際廠商與醫療研究人員提供與提出評估的流程與做法，但仍有些技術難題亟待被解決。Most of the symptoms of the disease will show slight differences in the variability of physiological organ operation and rhythm. However, although the monitoring and analysis of physiological signals have been provided by many international manufacturers and medical researchers, the procedures and practices for evaluation are still Some technical problems need to be solved.
以心電圖為例，目前對於心臟機能的檢測多以短時間的心電圖分析為主，但由於許多病症在短時間的心電圖上並無法顯示異常，因此近年已有研究人員針對長時間的心電圖訊號分析，發展了以多元尺度(multi-scale)觀點分析心臟律動複雜度(complexity)為主的生理參數指標，並在研究中證實此類指標確實能反應心臟機能的健康狀態。多元尺度生理參數指標的計算上較傳統時頻域的統計分析複雜，特別是以熵(entropy)為出發點的多元尺度熵(multi-scale entropy,MSE)，在醫學研究上已證明其有效性。Taking ECG as an example, the current detection of cardiac function is mainly based on short-term electrocardiogram analysis. However, since many diseases cannot display abnormalities on short-term electrocardiogram, researchers have analyzed the long-term ECG signal in recent years. A physiological parameter index based on multi-scale perspective analysis of heart rhythm complexity was developed, and it was confirmed in the study that such indicators can indeed reflect the health status of cardiac function. The calculation of multi-scale physiological parameters is more complicated than the statistical analysis of the traditional time-frequency domain, especially the multi-scale entropy (MSE) based on entropy, which has been proved to be effective in medical research.
長時間心電圖分析雖然提供了較完整的個案生理資訊，但相對的系統也必需有更大的空間以儲存長時間心電圖資料，如何設計出新的機制能在計算長時間心電圖生理參數指標時同時以有效率的方式儲存心電圖資訊，是長時間心電圖分析中的挑戰之一。Although long-term ECG analysis provides more complete case physiological information, the relative system must have more space to store long-term ECG data. How to design a new mechanism can calculate the long-term ECG physiological parameter index at the same time Efficient storage of ECG information is one of the challenges in long-term ECG analysis.
以多元尺度為出發點所發展的長時間生理參數指標，能呈現長時間範圍內的個案生理狀態，而此生理狀態的差異是無法透過短時間生理訊號所分析得到的；但由於計算時間相當冗長，因此目前仍較侷限於個案發病後的病症解讀與研究，未能將之應用於個案生理狀態的監控與預警。由此可見，如何提供此類多元尺度生理參數指標以盡可能即時監控與評估個案生理狀態，是臨床上個案生理監控相當重要的議題。The long-term physiological parameter index developed from the multi-scale is able to present the physiological state of the case within a long period of time, and the difference in the physiological state cannot be analyzed by the short-term physiological signal; however, since the calculation time is rather lengthy, Therefore, it is still limited to the interpretation and research of the disease after the onset of the case, and it has not been applied to the monitoring and early warning of the physiological state of the case. It can be seen that how to provide such multi-scale physiological parameter indicators to monitor and evaluate the physiological state of the case as soon as possible is a very important issue in clinical case monitoring.
本揭露提供一種運算生理參數指標的系統、方法、記錄媒體及電腦程式產品。The present disclosure provides a system, method, recording medium and computer program product for computing physiological parameter indicators.
本揭露之一實施例提出一種運算生理參數指標的方法，適用於電子裝置。此方法係將生理資料序列切割為多個視窗，其中各個視窗內包括生理資料序列中的一個資料區段。分析各個視窗內的資料區段，獲得能代表此資料區段之資料特性的元資料(metadata)。利用其中一個視窗的對應的元資料更新包含截至前一視窗為止的所有資料區段的資料特性的元資料，得到包含截至目前視窗為止的所有資料區段的資料特性的元資料。並利用更新後的元資料運算生理參數指標。One embodiment of the present disclosure proposes a method for calculating a physiological parameter index, which is applicable to an electronic device. The method cuts the physiological data sequence into a plurality of windows, wherein each window includes a data segment in the physiological data sequence. The data sections in each window are analyzed to obtain metadata that can represent the data characteristics of the data section. The metadata of the data characteristics of all the data sections up to the previous window is updated by using the corresponding metadata of one of the windows, and the metadata of the data characteristics of all the data sections up to the current window is obtained. And use the updated meta data to calculate physiological parameter indicators.
本揭露之一實施例提出一種運算生理參數指標的系統，其包括轉換器及電腦系統。其中，轉換器係用以檢測生理資料序列。電腦系統包括傳輸介面、至少一儲存媒體及處理器。其中，傳輸介面係連接轉換器，用以接收生理資料序列。至少一儲存媒體係用以儲存生理資料序列。處理器係耦接傳輸介面及該至少一儲存媒體，用以切割生理資料序列為多個視窗，分析各個視窗內生理資料序列中的資料區段，獲得能代表此資料區段之資料特性的元資料，並利用其中一個視窗對應的元資料更新包含截至前一個視窗為止的所有資料區段的資料特性的元資料，得到包含截至目前視窗為止的所有資料區段的資料特性的元資料，以及利用更新後的元資料運算生理參數指標。One embodiment of the present disclosure provides a system for computing physiological parameter indicators, including a converter and a computer system. The converter is used to detect a physiological data sequence. The computer system includes a transmission interface, at least one storage medium, and a processor. The transmission interface is connected to the converter for receiving the physiological data sequence. At least one storage medium is used to store a sequence of physiological data. The processor is coupled to the transmission interface and the at least one storage medium for cutting the physiological data sequence into a plurality of windows, analyzing the data segments in the physiological data sequence in each window, and obtaining a element capable of representing the data characteristics of the data segment. Data, and using the metadata corresponding to one of the windows to update the metadata of the data characteristics of all the data sections up to the previous window, obtain the metadata of the data characteristics of all the data sections up to the current window, and utilize the metadata The updated meta-data is used to calculate physiological parameter indicators.
本揭露之一實施例提出一種內儲程式之電腦可讀取記錄媒體，當電腦載入該程式並執行後，可完成如上所述之方法。One embodiment of the present disclosure provides a computer readable recording medium having a built-in program, and when the computer loads the program and executes it, the method as described above can be completed.
本揭露之一實施例提出一種運算生理參數指標電腦程式產品，當電腦載入該電腦程式並執行後，可完成如上所述之方法。One embodiment of the present disclosure provides a computer program product for computing physiological parameter indicators. When the computer loads the computer program and executes it, the method as described above can be completed.
為讓本發明之上述特徵和優點能更明顯易懂，下文特舉實施例，並配合所附圖式作詳細說明如下。The above described features and advantages of the present invention will be more apparent from the following description.
本揭露在長時間生理資料分析上，提出了計算生理參數指標的方法，利用將陸續進入系統的生理資料以視窗(window)概念分割出相對較短的生理資料區段，以避免批次運算時大量資料所造成的效率低落問題。此外，針對長時間生理資料的儲存空間問題，本揭露實施例也提供了利用元資料(metadata)特性來取代原本所儲存的原始生理資料，此方式在每個視窗進入時，更新元資料的資訊以描述過往所有資料的特性，原則上節省儲存成本。最後，本揭露實施例提出了以有系統的資料結構與元資料的資訊結合，藉以推算出長時間生理參數指標的做法。透過上述三項處理，本揭露能讓系統提供長時間生理參數指標，特別是多元尺度分析中運算最繁複的多元尺度熵(multi-scale entropy,MSE)指標，在本揭露實施例中同樣能提供醫護人員數據，使得長時間生理狀態的監控與分析能更廣泛地在臨床上被採用。In the long-term physiological data analysis, the present invention proposes a method for calculating physiological parameter indexes, and uses the physiological data that will enter the system successively to segment a relatively short physiological data segment by the window concept to avoid batch calculation. The inefficiency caused by a large amount of data. In addition, for the storage space problem of long-term physiological data, the disclosed embodiment also provides the use of meta-data characteristics to replace the original stored physiological data, which updates the metadata information when each window enters. In principle, the storage cost is saved by describing the characteristics of all previous data. Finally, the disclosed embodiment proposes a method of combining the information of a systematic data structure with the metadata to derive a long-term physiological parameter index. Through the above three treatments, the disclosure can provide the system with long-term physiological parameter indicators, especially the most complicated multi-scale entropy (MSE) index in multi-scale analysis, which can also be provided in the disclosed embodiment. Health care personnel data enables the monitoring and analysis of long-term physiological status to be more widely used clinically.
以下實施例以心電圖為例說明，但不限於只應用於心電圖。The following embodiments illustrate an electrocardiogram as an example, but are not limited to being applied only to an electrocardiogram.
圖1是依照本發明一實施例所繪示的運算生理參數指標方法的流程圖。請參照圖1，本實施例的生理參數指標的運算適用於具運算能力的各種電子裝置，其主要分為資料序列的分割、元資料的計算、元資料的累計與儲存及生理參數指標的計算等四個步驟，茲分述如下：在步驟S102中，由電子裝置接收一陸續進入裝置的生理資料序列，並將其切割為多個視窗，其中各個視窗內包括此生理資料序列中的一個資料區段。詳言之，本實施例所提出的資料序列分割，例如是將心電圖訊號中的RRI(R-R interval,RRI)、PRI(P-R interval,PRI)等能代表每次心跳週期資訊的序列，依據固定的時間長度(如半個小時或一個小時等)或資料長度(如5,000個RRI資料點或10,000個RRI資料點等)定義視窗大小(size)，將原始的長時間資料序列(如24小時的RRI序列)切割為數個不相互交疊(non-overlap)的資料區段，後續處理即針對每個資料區段進行運算。FIG. 1 is a flow chart of a method for calculating a physiological parameter index according to an embodiment of the invention. Referring to FIG. 1 , the calculation of the physiological parameter index of the present embodiment is applicable to various electronic devices with computing power, and is mainly divided into data sequence segmentation, meta data calculation, metadata accumulation and storage, and physiological parameter index calculation. The four steps are as follows: in step S102, the electronic device receives a physiological data sequence that successively enters the device, and cuts it into a plurality of windows, wherein each window includes one of the physiological data sequences. Section. In detail, the data sequence segmentation proposed in this embodiment is, for example, a sequence of RRI (RR interval, RRI), PRI (PR interval, PRI), etc. in the electrocardiogram signal, which can represent each heartbeat cycle information, according to a fixed The length of time (such as half an hour or an hour, etc.) or the length of the data (such as 5,000 RRI data points or 10,000 RRI data points, etc.) defines the size of the window, and the original long-term data sequence (such as 24-hour RRI) The sequence is cut into a number of non-overlapped data segments, and subsequent processing is performed for each data segment.
需說明的是，上述的生理資料序列係以心電圖特徵參數的資料序列為例，此心電圖特徵參數包括以時間觀點所量測到的心電圖中相鄰心跳的R波與R波間的時間長度(R-R interval,RRI)、單一心跳間期中的P波與R波的區間長度(P-R interval,PRI)、QRS波組時間長度(QRS duration)、S波與T波間的區段時間長度(ST segment)，以空間觀點所量測到的相鄰心跳間的P波、R波、S波、T波電位變化的差量，以及以型態觀點所量測到的相鄰心電圖間的型態差異的變化量或相似度其中之一。而除了心電圖特徵參數的資料序列之外，本實施例的方法亦適用於其他的生理資料序列，例如腦電波圖特徵參數、呼吸訊號或血氧濃度訊號的資料序列也可採用本實施例的做法運算對應的生理參數指標。It should be noted that the above physiological data sequence is taken as an example of a data sequence of an electrocardiographic characteristic parameter, and the electrocardiographic characteristic parameter includes a time length between an R wave and an R wave of an adjacent heartbeat in an electrocardiogram measured from a time point of view (RR). Interval, RRI), the interval length of the P wave and the R wave in a single heartbeat interval (PRI), the length of the QRS duration (QRS duration), and the length of the segment between the S wave and the T wave (ST segment). The difference between the P wave, R wave, S wave, and T wave potential changes between adjacent heartbeats measured from the spatial point of view, and the change in the type difference between adjacent ECGs measured by the type view One of quantity or similarity. In addition to the data sequence of the ECG characteristic parameter, the method of the embodiment is also applicable to other physiological data sequences, for example, the data sequence of the electroencephalogram characteristic parameter, the respiratory signal or the blood oxygen concentration signal can also adopt the practice of the embodiment. Calculate the corresponding physiological parameter indicator.
在步驟S104中，由電子裝置分析各個視窗內的資料區段，以獲得能代表資料區段之資料特性的元資料(metadata)。詳言之，依據所欲計算的生理參數指標之運算性質，本實施例可分析計算生理參數指標必要的元資料以及其計算方式，此元資料並能夠用以計算生理參數指標。In step S104, the data sections in the respective windows are analyzed by the electronic device to obtain metadata capable of representing the data characteristics of the data sections. In detail, according to the operational properties of the physiological parameter index to be calculated, the present embodiment can analyze and calculate the necessary metadata of the physiological parameter index and the calculation method thereof, and the metadata can be used to calculate the physiological parameter index.
需說明的是，上述元資料例如是用以代表資料特性的統計描述、資料結構特性、趨勢資訊或資料亂度量測值。其中，所述的統計描述包括平均數、標準差、眾數、中位數、偏態係數、峰態係數或機率分佈之參數；所述的資料結構特性包括資料直方圖分組或計數；所述的趨勢資訊包括迴歸係數或多項式係數；所述的資料亂度包括熵或時間非對稱性係數其中之一，在此均不設限。It should be noted that the above meta-data is, for example, a statistical description, a data structure characteristic, a trend information, or a data metric measurement value representing the characteristics of the data. Wherein, the statistical description includes parameters of mean, standard deviation, mode, median, skewness coefficient, kurtosis coefficient or probability distribution; the data structure characteristics include data histogram grouping or counting; The trend information includes regression coefficients or polynomial coefficients; the data disorder includes one of entropy or time asymmetry coefficients, and there is no limit here.
在步驟S106，由電子裝置利用其中一個視窗對應的元資料更新包含截至前一視窗為止的所有資料區段的資料特性的元資料，得到包含截至目前視窗為止的所有資料區段的資料特性的元資料。詳言之，隨著陸續進入系統的資料序列，本實施例提供一種元資料的更新方法，使得所更新後的元資料能夠代表截至目前為止已進入系統的資料序列之整體特性。需說明的是，對於此元資料的儲存，本實施例特別採用多維稀疏矩陣或樹狀資料結構來記錄，以達到節省儲存空間，其實施方式詳述於後續實施例。In step S106, the electronic device updates the metadata of the data characteristics of all the data sections up to the previous window by using the metadata corresponding to one of the windows to obtain the metadata of the data characteristics of all the data sections up to the current window. data. In detail, with the data sequence entering the system one after another, the embodiment provides a method for updating the metadata so that the updated metadata can represent the overall characteristics of the data sequence that has entered the system so far. It should be noted that, for the storage of the metadata, the embodiment is specifically recorded by using a multi-dimensional sparse matrix or a tree data structure to save storage space, and an embodiment thereof is described in detail in subsequent embodiments.
在步驟S108，由電子裝置利用更新後的元資料運算生理參數指標。詳言之，在將每個時間區段的元資料更新後，此更新後的元資料即可用以計算生理參數指標。由於元資料與原始的生理資料序列不同，因此計算生理參數指標的方式也與傳統方法相異。為計算出生理參數指標，本實施例提供額外的資料處理架構以進行運算，其實施方式詳述於後續實施例。In step S108, the physiological parameter index is calculated by the electronic device using the updated metadata. In detail, after updating the metadata of each time segment, the updated metadata can be used to calculate physiological parameter indicators. Since the meta-data is different from the original physiological data sequence, the way to calculate the physiological parameter index is also different from the traditional method. To calculate physiological parameter indicators, this embodiment provides an additional data processing architecture to perform the operations, the implementation of which is detailed in subsequent embodiments.
圖2是依照本發明一實施例所繪示的運算生理參數指標方法的示意圖。請參照圖2，本實施例係將生理資料序列20依據固定的時間長度(T 1 、T 2 、…、T k )切割為多個較短時間長度的資料區段(包括RRI資料區段1、RRI資料區段2、…、RRI資料區段k )。而在對這些資料區段進行分析的過程中，本實施例會先執行多元尺度分析中的粗粒化(coarse-graining)程序，再利用特定的資料結構計算與儲存能代表各個資料區段之資料特性的元資料(包括元資料1、元資料2、…、元資料k )。其中，所述的元資料將隨著每個進入的資料區段不斷更新，以獲得能代表全部長時間心電圖特性的元資料，最後再進行近似熵或樣本熵的計算，從而獲得所需的長時間生理參數指標22，並完成多元尺度熵的計算程序。FIG. 2 is a schematic diagram of a method for calculating a physiological parameter index according to an embodiment of the invention. Referring to FIG. 2, in this embodiment, the physiological data sequence 20 is cut into a plurality of short-length data sections according to a fixed time length ( T 1 , T 2 , . . . , T k ) (including the RRI data section 1). , RRI data section 2, ..., RRI data section k ). In the process of analyzing these data sections, the present embodiment first performs a coarse-graining procedure in multi-scale analysis, and then uses a specific data structure to calculate and store data representing each data section. Metadata of characteristics (including metadata 1, metadata 2, ..., metadata k ). Wherein, the meta-data will be continuously updated with each incoming data section to obtain meta-data capable of representing all long-term ECG characteristics, and finally the approximate entropy or sample entropy calculation is performed to obtain the required length. Time physiological parameter index 22, and complete the calculation procedure of multi-scale entropy.
需說明的是，上述的粗粒化程序例如是使用多個尺度(scale)分別對各個視窗內的資料區段進行運算，以獲得各個尺度下的資料序列，並用以計算能代表資料區段之資料特性的元資料。其中，在使用上述尺度其中之一對資料區段執行粗粒化程序時，例如是以所使用之尺度為單位，依序選擇資料區段中的多筆資料，並計算所選擇資料的平均值，而用以作為該尺度下的資料序列中的一筆資料。It should be noted that the above-mentioned coarse granulation program is used, for example, to calculate data segments in each window by using a plurality of scales to obtain data sequences at various scales, and to calculate data segments. Metadata of data characteristics. Wherein, when the coarsening procedure is performed on the data section using one of the above-mentioned scales, for example, the plurality of data in the data section are sequentially selected in units of scales used, and the average value of the selected data is calculated. And used as a piece of data in the data sequence at this scale.
舉例來說，圖3(a)及圖3(b)是依照本發明一實施例所繪示的粗粒化程序的範例。請同時參照圖3(a)及圖3(b)，對於給定的資料序列X ={x i }，其中1 i N ，對此資料序列X 進行粗粒化程序後的資料序列可由以下公式獲得：For example, Figures 3(a) and 3(b) are examples of coarse graining procedures in accordance with an embodiment of the present invention. Please refer to FIG. 3(a) and FIG. 3(b) simultaneously, for a given data sequence X = { x i }, where 1 i N , the data sequence after the coarsening process of the data sequence X It can be obtained by the following formula:
其中，N 為資料序列X 所包括的資料總筆數，τ 為所選擇使用的粗粒化尺度。由圖3(a)可知，在尺度τ =2時，即是以2為單位依序選擇資料區段X 中的2筆資料，例如(x 1 ,x 2 )、(x 3 ,x 4 )、(x 5 ,x 6 )、…、(x i ,x i+1 )…，並計算所選擇資料的平均值，而用以作為該尺度下的資料序列中的資料，最終獲得粗粒化程序後的資料序列V τ =V 2 =(v 1 ,v 2 ,v 3 ,…)。同理，由圖3(b)可知，在尺度τ =3時，是以3為單位依序選擇資料區段X 中的3筆資料，並計算所選擇資料的平均值，而用以作為該尺度下的資料序列中的資料，最終獲得粗粒化程序後的資料序列V 3 =。以實際數字為例，若原始的資料序列X =(26,28,30,26,26,27,25)，則經過粗粒化程序(τ =2)後的資料序列V 2 =(27,28,26.5)。Where N is the total number of data included in the data sequence X , and τ is the coarse granulation scale selected. As can be seen from Fig. 3(a), when the scale τ = 2, two data in the data segment X are sequentially selected in units of 2, for example, ( x 1 , x 2 ), ( x 3 , x 4 ) , ( x 5 , x 6 ), ..., ( x i , x i+1 )..., and calculate the average value of the selected data, and use it as data in the data sequence at the scale to finally obtain coarse graining The data sequence after the program V τ = V 2 = ( v 1 , v 2 , v 3 , ...). Similarly, as shown in Fig. 3(b), when the scale τ = 3, three data in the data segment X are sequentially selected in units of 3, and the average value of the selected data is calculated and used as the The data in the data sequence under the scale, finally obtain the data sequence after the coarse granulation procedure V 3 = . Taking the actual number as an example, if the original data sequence X = (26, 28, 30, 26, 26, 27, 25), then the data sequence V 2 = (27, after the coarse graining procedure ( τ = 2). 28, 26.5).
關於元資料的計算及更新，圖4是依照本發明一實施例所繪示的計算多元尺度熵之資料結構所對應的直方圖。請參照圖4，直方圖40係對應於本實施例針對多元尺度熵中觀察樣本維度設定為m =2所推演的元資料進行運算時所採用的資料結構。其中，由於近似熵或樣本熵的計算需要統計每個樣本點與其他樣本點的相互關係(例如樣本值的差量)，因此本實施例係利用多維直方圖的處理架構，將每個樣本點(以本實施例而言皆是二維向量)依據各個維度樣本值大小順序排列並計算各種組合(二維向量)出現的次數，進而統計整理成二維統計表。當觀察樣本維度設定為m=3時，此處理方式可統計整理成三維統計表。For the calculation and update of the metadata, FIG. 4 is a histogram corresponding to the data structure for calculating the multi-scale entropy according to an embodiment of the invention. Referring to FIG. 4, the histogram 40 corresponds to the data structure used in the calculation of the meta-data derived from the observation sample dimension set to m =2 in the multi-scale entropy. Wherein, since the calculation of the approximate entropy or the sample entropy needs to calculate the correlation between each sample point and other sample points (for example, the difference of the sample values), the present embodiment uses the processing architecture of the multi-dimensional histogram to set each sample point. (In the present embodiment, both are two-dimensional vectors) Arrange and calculate the number of occurrences of various combinations (two-dimensional vectors) according to the size of each dimension sample value, and then statistically organize them into a two-dimensional statistical table. When the observed sample dimension is set to m=3, this processing mode can be statistically organized into a three-dimensional statistical table.
雖然每次進入的資料區段的樣本點數目可能高達數千/萬筆，但在第一維樣本數值條件的限制下，第二維樣本數值的分佈廣度極其有限，此現象對心律周期的生理分析而言也甚合理，因為相鄰兩次心跳的周期差異並不會出現巨大的改變。上述現象在限制第一維與第二維樣本數值下的第三維分佈(m =3設定時)更為明顯(即第三維樣本數值的變動幅度也有限)。Although the number of sample points in each data segment may be as high as several thousand/10,000 strokes, the distribution of the second-dimensional sample values is extremely limited under the constraints of the first-dimensional sample numerical conditions. This phenomenon is physiological for the heart rhythm cycle. It is also reasonable to analyze, because the cycle difference between two adjacent heartbeats does not change dramatically. The above phenomenon is more obvious when limiting the third-dimensional distribution under the first-dimensional and second-dimensional sample values (when m = 3 is set) (that is, the variation of the value of the third-dimensional sample is also limited).
依上述所觀察之現象，在二維統計表或三維統計表中，每個單位(cell)有值的機率將遠小於每個單位無值的機率(即該二維或三維的組合在資料中未曾出現)。若要將二維統計表或三維統計表完整記錄，則將會浪費許多不必要的儲存空間。有鑑於此，也可以採用多維稀疏矩陣(sparse matrix)來記錄統計表，藉以節省儲存空間。According to the phenomenon observed above, in a two-dimensional statistical table or a three-dimensional statistical table, the probability that each unit has a value will be much smaller than the probability that each unit has no value (ie, the two-dimensional or three-dimensional combination is in the data. Never appeared). To fully record a 2D or 3D statistic, you will waste a lot of unnecessary storage space. In view of this, a multi-dimensional sparse matrix can also be used to record statistical tables, thereby saving storage space.
詳言之，對於由生理資料序列切割出來的各個視窗的各尺度下的資料序列，本發明實施例將其中一個視窗對應的元資料記錄於多維稀疏矩陣，再依序將其他視窗對應的元資料累加到同一個多維稀疏矩陣，使得此多維稀疏矩陣包含截至目前視窗為止的所有資料區段的資料特性的元資料。In detail, for the data sequence at each scale of each window cut out by the physiological data sequence, the embodiment of the present invention records the meta data corresponding to one window in the multi-dimensional sparse matrix, and sequentially stores the metadata corresponding to the other windows. The accumulation is added to the same multi-dimensional sparse matrix such that the multi-dimensional sparse matrix contains metadata of the data characteristics of all data sections up to the current window.
上述記錄元資料的步驟例如是先將其中一個視窗對應的元資料所包含的每個向量組合的計數值記錄於多維稀疏矩陣，然後再依序將其他視窗對應的元資料所包含的每個向量組合的計數值累加於此多維稀疏矩陣中已記錄的向量組合的計數值上。其中，上述累加到多維稀疏矩陣的向量組合與多維稀疏矩陣中原本的向量組合之間可能存在相同及相異的部分，對於第二向量組合與第一向量組合相同的部分，將兩者的計數值進行累加；反之，對於第二向量組合與第一向量組合相異的部分，在多維稀疏矩陣中並未有對應的第一向量組合，因此本發明實施例需根據第二向量組合對於多維稀疏矩陣中所有向量組合的相對位置，適度擴大多維稀疏矩陣的尺寸，以將第二向量組合納入多維稀疏矩陣中，並用以做為新增的向量組合。The step of recording the metadata includes, for example, first recording the count value of each vector combination included in the metadata corresponding to one of the windows in the multi-dimensional sparse matrix, and then sequentially instructing each vector included in the metadata corresponding to the other windows. The combined count value is accumulated over the count value of the recorded vector combination in the multi-dimensional sparse matrix. Wherein, the above-mentioned vector combination of the vector combination added to the multi-dimensional sparse matrix and the original vector combination in the multi-dimensional sparse matrix may have the same and different parts, and for the second vector combination and the same part of the first vector combination, the two The value is accumulated; otherwise, for the portion where the second vector combination is different from the first vector combination, there is no corresponding first vector combination in the multi-dimensional sparse matrix, so the embodiment of the present invention needs to be multi-dimensional sparse according to the second vector combination. The relative positions of all vector combinations in the matrix moderately expand the size of the multi-dimensional sparse matrix to incorporate the second vector combination into the multi-dimensional sparse matrix and use it as a new vector combination.
舉例來說，圖5是依照本發明一實施例所繪示的使用稀疏矩陣儲存並更新元資料的範例。請參照圖5，本實施例係說明在計算多元尺度熵的尺度m =2條件下，元資料的記錄及更新方式。由圖中可知，本實施例會先以多維稀疏矩陣所記錄更新至第t 時間區段(對應視窗t )的元資料(包含前t 個時間區段的資訊)與第t +1時間區段(對應視窗t +1)所計算出的元資料轉換成完整的資訊矩陣(full information matrix)，進行元資料的更新，並將更新後資訊矩陣中的元資料以多維稀疏矩陣的方式儲存以作為更新至第t +1時間區段的元資料。依此方式，由元資料所轉換的資訊矩陣將依據隨時間順序進入的每個資料區段不斷更新資訊，而資訊矩陣的大小在每次更新的過程中有可能會增長或維持不變。For example, FIG. 5 is an example of storing and updating metadata using a sparse matrix according to an embodiment of the invention. Referring to FIG. 5, this embodiment illustrates the manner in which metadata is recorded and updated under the condition that the scale m =2 of the multivariate scale entropy is calculated. Seen from the figure, the present embodiment would in a first multi-dimensional sparse matrix metadata (information containing time zone before t) to update the record of the time interval t (corresponding to the windows t) t + 1'd and second time segments ( The metadata calculated corresponding to the window t +1) is converted into a complete information matrix, the metadata is updated, and the metadata in the updated information matrix is stored as a multi-dimensional sparse matrix as an update. Metadata to the t +1th time zone. In this way, the information matrix transformed by the metadata will continuously update the information according to each data segment that enters in time sequence, and the size of the information matrix may grow or remain unchanged during each update.
圖6是依照本發明一實施例所繪示的利用不斷更新的元資料計算多元尺度熵的示意圖。請參照圖6，由於在元資料更新時不斷統計出每種向量組合出現的次數，因此本實施例係依據元資料所轉換出的資訊矩陣60，設定多元尺度熵中所定義的差量上限為r=0.15×SD ，其中SD 為截至目前時間點所有RRI資料的標準差。藉此，可框出與特定向量ω ( p , q ) 之差量小於此差量上限r的區塊62，而將區塊62範圍中所有向量組合的計數值相加後，即可得到與向量ω ( p , q ) 相近的計數值總和。依據目標向量ω ( p , q ) 在目前累計資料中出現的次數和距ω ( p , q ) 在r範圍內的總資料數，並對資訊矩陣60中的每個位置進行此動作，可得出運算近似熵或樣本熵所需的所有資訊。考量多元尺度下的多組資料序列，執行上述運作可計算出多元尺度熵參數指標。FIG. 6 is a schematic diagram of calculating multi-scale entropy using continuously updated metadata according to an embodiment of the invention. Referring to FIG. 6 , since the number of occurrences of each vector combination is continuously counted when the metadata is updated, the present embodiment sets the upper limit of the difference defined in the multi-scale entropy according to the information matrix 60 converted by the metadata. r = 0.15 x SD , where SD is the standard deviation of all RRI data up to the current time point. Thereby, the block 62 which is different from the specific vector ω ( p , q ) by less than the upper limit r of the difference can be framed, and the count values of all the vector combinations in the range of the block 62 can be added to obtain the The sum of the count values of the vectors ω ( p , q ) are similar. According to the number of times the target vector ω ( p , q ) appears in the current accumulated data and the total number of data in the range of distance ω ( p , q ) , and the action is performed for each position in the information matrix 60, All the information needed to approximate the entropy or sample entropy. Considering multiple sets of data sequences at multiple scales, the multi-scale entropy parameter can be calculated by performing the above operations.
除了使用多維稀疏矩陣記錄元資料的方式之外，本發明一實施例還提供另一種利用樹狀資料結構記錄元資料的方式。In addition to the manner in which metadata is recorded using a multi-dimensional sparse matrix, an embodiment of the present invention provides another way of recording metadata using a tree-like data structure.
詳言之，例如是針對各個視窗的各個尺度下的資料序列，將其中一個視窗對應的元資料記錄於樹狀資料結構，然後再將其他視窗對應的元資料依序加入此樹狀資料結構，使得此樹狀資料結構包含截至目前視窗為止的所有資料區段之資料特性的元資料。In detail, for example, for each data sequence of each window, the metadata corresponding to one window is recorded in the tree data structure, and then the metadata corresponding to other windows is sequentially added to the tree data structure. This tree data structure is made to contain metadata about the data characteristics of all data sections up to the current window.
舉例來說，圖7(a)、圖7(b)及圖7(c)是依照本發明一實施例所繪示的利用樹狀資料結構記錄元資料，並據以運算生理參數指標的範例。其中，本實施例係將資料區段X =(3,10,19,23,30,37,45,59,62,70,80,89,95,98)中的樣本點記錄於二元樹(binary tree)資料結構(如圖7(a)所示)，或是記錄於一維樹(1D tree)資料結構(如圖7(b)所示)。而在運算生理參數指標時，則是先設定多元尺度熵中所定義的差量上限，然後針對樹狀資料結構中的特定樣本點，搜尋樹狀資料結構中與此樣本點之差量小於差量上限的範圍，最後則計算此範圍中所有向量組合的計數值總和，以做為計算近似熵/樣本熵生理參數指標所必需的資訊。例如在圖7(c)中，即將差量上限設定為30，而針對樣本點x =45搜尋圖7(b)的樹狀資料結構中與此樣本點之差量小於差量上限的範圍(即x =15~75)，最後則計算此範圍中所有樣本點的計數值總和，以做為計算近似熵/樣本熵生理參數指標所必需的資訊。For example, FIG. 7(a), FIG. 7(b), and FIG. 7(c) are diagrams illustrating the use of a tree-like data structure to record metadata and calculate physiological parameter indicators according to an embodiment of the invention. . In this embodiment, the sample points in the data segment X = (3, 10, 19, 23, 30, 37, 45, 59, 62, 70, 80, 89, 95, 98) are recorded in the binary tree. (binary tree) data structure (as shown in Figure 7 (a)), or recorded in the 1D tree (1D tree) data structure (as shown in Figure 7 (b)). In the calculation of the physiological parameter index, the upper limit of the difference defined in the multivariate scale entropy is first set, and then the difference between the searched tree data structure and the sample point is less than the difference for the specific sample point in the tree data structure. The range of the upper limit, and finally the sum of the count values of all vector combinations in this range is calculated as the information necessary to calculate the approximate entropy/sample entropy physiological parameter. For example, in FIG. 7(c), the upper limit of the difference is set to 30, and for the sample point x =45, the range of the difference between the sample point and the sample point in the tree data structure of FIG. 7(b) is less than the upper limit of the difference ( That is, x = 15~75), and finally, the sum of the count values of all the sample points in the range is calculated as the information necessary for calculating the approximate entropy/sample entropy physiological parameter index.
根據一些實施例，第8圖為一運算生理參數指標系統的功能區塊圖，此系統可用於執行揭露在第1圖至第7圖中的方法。In accordance with some embodiments, FIG. 8 is a functional block diagram of an operational physiological parameter indicator system that can be used to perform the methods disclosed in FIGS. 1 through 7.
運算生理參數指標系統800包括一電腦系統810。電腦系統810包括了電力上與至少一儲存媒體812直接連接的一處理器814。為了使電腦像一信號分析器一樣執行受測生理訊號的生理參數指標計算與分析，可配置處理器814以執行或中止編譯於至少一儲存媒體812之電腦程式碼。The operational physiological parameter indicator system 800 includes a computer system 810. Computer system 810 includes a processor 814 that is electrically coupled directly to at least one storage medium 812. In order for the computer to perform the calculation and analysis of the physiological parameter indicators of the measured physiological signals as a signal analyzer, the processor 814 can be configured to execute or abort the computer code compiled on the at least one storage medium 812.
在一些實施例中，處理器814為一中央處理單元(central processing unit,CPU)、一多元處理器、一分散式處理系統及/或任一適合的處理單元。在至少一實施例中，處理器814可從至少一儲存媒體812中取得像是心電圖訊號的生理訊號、預定標準樣版及/或其他資訊。In some embodiments, processor 814 is a central processing unit (CPU), a multi-processor, a decentralized processing system, and/or any suitable processing unit. In at least one embodiment, the processor 814 can obtain physiological signals, predetermined standard patterns, and/or other information, such as an electrocardiogram signal, from the at least one storage medium 812.
在一些實施例中，至少一儲存媒體812為一電子、磁性、光學、電磁、紅外線及/或一半導體系統(或儀器或裝置)。舉例來說，至少一儲存媒體812包括一半導體或固態記憶體、一磁帶、一可攜式電腦磁片、一隨機存取記憶體(random access memory,RAM)、一唯讀記憶體(read-only memory,ROM)、一硬式磁碟及/或一光學磁碟。在一些使用光學磁碟的實施例中，至少一儲存媒體812包括一唯讀式光碟(compact disc read-only memory,CD-ROM)、一可重複讀寫光碟(compact disc rewritable,CD-RW)及/或一數位影音光碟(digital video disk,DVD)等。In some embodiments, the at least one storage medium 812 is an electronic, magnetic, optical, electromagnetic, infrared, and/or a semiconductor system (or instrument or device). For example, at least one storage medium 812 includes a semiconductor or solid state memory, a magnetic tape, a portable computer magnetic disk, a random access memory (RAM), and a read-only memory (read- Only memory, ROM), a hard disk and/or an optical disk. In some embodiments using an optical disk, the at least one storage medium 812 includes a compact disc read-only memory (CD-ROM) and a compact disc rewritable (CD-RW). And / or a digital video disk (DVD) and so on.
此外，電腦系統810包括一輸入/輸出介面816及一顯示器818。輸入/輸出介面816與處理器814直接連接，並且為了執行在第1圖至第7圖中所描述的方法，可允許一操作者或醫療照護專業人員操作電腦系統810。第1圖至第7圖中所描述之方法的操作狀況，能藉圖形使用者介面(graphical user interface,GUI)顯現於顯示器818中。輸入/輸出介面816及顯示器818以人機互動的方式允許一操作者操作電腦系統810。In addition, computer system 810 includes an input/output interface 816 and a display 818. The input/output interface 816 is directly coupled to the processor 814 and may allow an operator or medical care professional to operate the computer system 810 in order to perform the methods described in FIGS. 1 through 7. The operational status of the method described in Figures 1 through 7 can be visualized in display 818 by a graphical user interface (GUI). Input/output interface 816 and display 818 allow an operator to operate computer system 810 in a human-machine interactive manner.
在一實施例，電腦系統810也包括直接連接至處理器814的一網路介面822。網路介面822允許電腦系統810能與和一網路830相連之一或更多電腦系統通信。網路介面822包括無線網路介面像是藍芽(BLUETOOTH)、無線相容認證(wireless fidelity,WIFI)、全球互通微波存取(worldwide interoperability for microwave access,WiMAX)、通用封包無線服務技術(general packet radio service,GPRS)、寬頻分碼多工(wide band code division multiple access,WCDMA)、有線網路介面像是乙太網路(ETHERNET)、通用序列匯流排(USB)或IEEE-1394。在一些實施例中，第1圖至第7圖之方法可執行在第8圖兩個或更多個電腦系統810中，例如心電圖訊號等生理訊號、預定標準樣版、及/或其他的資訊能經由網路830在不同的電腦系統中所交換。In one embodiment, computer system 810 also includes a network interface 822 that is directly coupled to processor 814. Network interface 822 allows computer system 810 to communicate with one or more computer systems connected to a network 830. The network interface 822 includes a wireless network interface such as BLUETOOTH, wireless fidelity (WIFI), worldwide interoperability for microwave access (WiMAX), and general packet wireless service technology (general). Packet radio service (GPRS), wide band code division multiple access (WCDMA), wired network interface such as Ethernet (ETHERNET), Universal Serial Bus (USB) or IEEE-1394. In some embodiments, the methods of FIGS. 1-7 can be performed in two or more computer systems 810 of FIG. 8, such as physiological signals such as electrocardiogram signals, predetermined standard templates, and/or other information. Can be exchanged over different networks of the computer system via the network 830.
在至少一實施例中，運算生理參數指標系統800更包含一轉換器840。轉換器840能用於觀察被檢測之生物個體/器官，以及轉換生物個體/器官的運動運作為一具代表性之訊號。在一分析心電圖信號的實施例中，轉換器840觀察被檢測的心臟並且轉換心臟肌肉運動運作為心電圖訊號。In at least one embodiment, the operational physiological parameter indicator system 800 further includes a transducer 840. The transducer 840 can be used to observe the organism/organ of the organism being tested, and to convert the movement of the organism/organ into a representative signal. In an embodiment where the electrocardiographic signal is analyzed, the transducer 840 observes the detected heart and converts the cardiac muscle motion to operate as an electrocardiogram signal.
電腦系統810更有一與轉換器840及處理器814直接連接的傳輸介面824。傳輸介面824能橋接轉換器840及處理器814，並將所取得的週期信號例如以離散時間訊號格式輸出。舉例來說，若轉換器840取得一心電圖訊號，則傳輸介面824從轉換器840接收心電圖訊號，並將心電圖信號以心電圖資料陣列格式輸出至處理器814中。在一些實施例中，轉換器840轉換下列生物個體現象的其中之一為電子信號：心跳、呼吸、心電圖、腦電波、血氧濃度等生理訊號、等。The computer system 810 further has a transmission interface 824 that is directly coupled to the converter 840 and the processor 814. The transmission interface 824 can bridge the converter 840 and the processor 814 and output the acquired periodic signals, for example, in a discrete time signal format. For example, if the converter 840 takes an ECG signal, the transmission interface 824 receives the ECG signal from the converter 840 and outputs the ECG signal to the processor 814 in an ECG data array format. In some embodiments, converter 840 converts one of the following biological individual phenomena to an electrical signal: heartbeat, respiration, electrocardiogram, brainwave, blood oxygen concentration, etc., physiological signals, and the like.
為了驗證本揭露之運算生理參數指標的方法相較於先前技術的改進，本案以三種不同的計算方式評估多元尺度熵的計算時間，包含多元尺度熵並未對資料做任何資料結構處理的原始做法、本案所提出的以有秩序的資料結構儲存元資料並直接以結構化批次處理方式計算多元尺度熵的做法，以及本案所提出的結構化線上計算多元尺度熵的做法。其中，在線上計算多元尺度熵的做法中，會再加以評估整個計算過程中所耗費的所有時間(包含在每個進入的時間區段中計算與更新元資料的時間以及計算多元尺度熵的時間)，以及操作人員實際等待系統運算出完整多元尺度熵的時間(包含一次的元資料運算與更新以及多元尺度熵的計算)，並分別表示為結構化線上計算(所有)做法與結構化線上計算(反應時間)做法。其中，本實施例所採用的生理資料序列為24小時的心電圖資料，其為經過自動化的R波特徵點偵測以及異位波的濾除後，再經由專業人員進行人工的校正，所得出的RRI序列資料。在下述的實施例中，視窗長度設定上，以固定資料量方式切割陸續進入系統的生理資料，並將每個視窗(時間區段)的資料大小設定為10,000筆。In order to verify the method of calculating the physiological parameter index of the present disclosure compared with the prior art improvement, the present case evaluates the calculation time of the multi-scale entropy in three different calculation manners, including the original practice that the multi-scale entropy does not perform any data structure processing on the data. The practice of storing multi-scale entropy in an orderly data structure and directly calculating the multi-scale entropy in structured batch processing, and the practice of calculating multi-scale entropy on the structured line proposed in this case. Among them, in the calculation of multi-scale entropy on the line, all the time spent in the whole calculation process (including the time of calculating and updating the metadata in each entering time segment and the time of calculating the multi-scale entropy) will be evaluated. ), and the time the operator actually waits for the system to compute the complete multi-scale entropy (including one-time metadata operations and updates and multi-scale entropy calculations), and represents the computational online (all) practices and structured online calculations, respectively. (Reaction time) practice. The physiological data sequence used in the embodiment is a 24-hour electrocardiogram data, which is obtained by automated R wave feature point detection and ectopic wave filtering, and then manually corrected by a professional. RRI sequence data. In the embodiment described below, the window length is set to cut the physiological data successively entering the system in a fixed amount of data, and the data size of each window (time zone) is set to 10,000.
圖9是依照一實施例所繪示的利用四種評估方式計算多元尺度熵在尺度設定為1的時間分析圖。從此圖可清楚看出原始未考量資料結構與運算效率的暴力做法在時間效率的表現上遠較於其他三種評估方式為差，特別在資料筆數超過30,000筆時，原始暴力做法所需的時間已超過其餘三種評估方法所需時間的百倍。從圖中也可驗証原始暴力做法計算方法的運算複雜度將隨資料筆數增加，而以指數倍數成長。FIG. 9 is a time analysis diagram for calculating a multi-scale entropy with a scale set to 1 using four evaluation methods according to an embodiment. It can be clearly seen from this figure that the violent practices of the original unconsidered data structure and computational efficiency are far worse in terms of time efficiency than the other three assessment methods, especially when the number of data exceeds 30,000, the time required for the original violent practice. It has exceeded 100 times the time required for the remaining three evaluation methods. It can also be verified from the figure that the computational complexity of the original violent practice calculation method will increase with the number of data and increase by exponential multiple.
圖10是依據圖9的實驗設定，將焦點放置於結構化方式計算多元尺度熵的比較。從圖中可發現，在資料量小於40,000時，利用結構化的批次計算方法比結構化線上計算方式的總運算時間略為有效率，但在總資料筆數大於40,000時，結構化線上計算(所有)做法的運算效率則優於暴力做法；若再與結構化線上計算(反應時間)做法的運算時間相比，則明顯看出所耗費的時間極短，當資料筆數為120,000時，僅需1.5秒的時間，約為結構化線上計算(所有)做法的二分之一以及約為結構化批次計算方法的三分之一時間，而此1.5秒的時間也是臨床醫療人員操作計算此生理參數指標時實際等待的時間。Figure 10 is a comparison of the multi-scale entropy calculated by placing the focus in a structured manner in accordance with the experimental setup of Figure 9. It can be seen from the figure that when the amount of data is less than 40,000, the structured batch calculation method is slightly more efficient than the total calculation time of the structured online calculation method, but when the total number of data is greater than 40,000, the structured online calculation ( All of the practices are more efficient than violent practices; if compared to the computational time of the structured online (reaction time) approach, it is clear that the time spent is extremely short, when the number of data is 120,000, only The 1.5 second time is about one-half of the computational (all) practice on the structured line and about one-third of the time of the structured batch calculation method, and this 1.5 second time is also the clinical medical staff's operation to calculate this physiology. The actual waiting time for the parameter indicator.
圖11是依照一實施例所繪示的利用四種評估方式計算多元尺度熵在尺度設定為1到20所耗用的總時間分析圖。與圖9呈現的趨勢相似，暴力做法的運算時間遠久於資料經結構化後再加以計算多元尺度熵的時間，此圖並可說明為何多元尺度熵雖在研究報告上有其效果但卻未能普及的原因，以資料筆數為120,000為例，原始計算方式需超過5,000秒的運算時間，此對於臨床評估應用上的即時性需求還有很大的進步空間。FIG. 11 is a diagram showing total time analysis of multivariate scale entropy used to calculate the scale from 1 to 20 using four evaluation methods according to an embodiment. Similar to the trend presented in Figure 9, the violent practice takes longer than the time after the data is structured to calculate the multivariate entropy. This graph shows why the multivariate entropy has an effect on the research report but it is not The reason why it can be popularized is that the original calculation method requires more than 5,000 seconds of calculation time, and there is still much room for improvement in the immediacy demand for clinical evaluation applications.
圖12是分析圖11中以資料結構化方式計算多元尺度熵的時間。與圖10結果有所差異的地方在於考量尺度1到20後，結構化線上計算(所有)做法所需的總計算時間皆高於暴力做法，此原因在於當scale大於3時，生理資料經粗粒化程序後資料筆數大為降低，直接以結構化的批次計算方式處理將比線上方式處理來得有效率，因為線上方式在每個視窗的資料處理中會額外有元資料的更新時間，造成總計算時間略高於批次運算方式的時間。但若以實際上臨床醫療人員的等待時間而言，結構化線上計算(反應時間)做法的計算時間則相似於圖10所展現的結果，結構化線上計算(反應時間)做法耗用時間相較於結構化的批次計算方法與結構化線上計算(所有)做法更為有效率，運算時間同樣僅需結構化線上計算(所有)做法的二分之一以及結構化的批次計算方法的約三分之一。由實施例中也可証實本案所提方法的可行性與計算效率性，透過序列資料學習的線上資料處理技術，提供長時間生理參數指標。Figure 12 is a graph for analyzing the time at which the multivariate scale entropy is calculated in a data structured manner in Figure 11. The difference from the results in Figure 10 is that after considering the scales 1 to 20, the total calculation time required for the calculation of (all) practices on the structured line is higher than the violent practice, because when the scale is greater than 3, the physiological data is coarse. After the granulation process, the number of data is greatly reduced. Directly processing in a structured batch calculation method will be more efficient than online processing, because the online method will additionally update the metadata in each window data processing. The total calculation time is slightly higher than the batch operation mode. However, in terms of the waiting time of the actual clinical staff, the calculation time of the structured online calculation (reaction time) is similar to the results shown in Figure 10, and the time spent on the structured online calculation (reaction time) is compared. The structured batch calculation method and the structured online calculation (all) are more efficient, and the computation time is only required to be one-half of the structured online calculation (all) and the structured batch calculation method. one third. The feasibility and computational efficiency of the proposed method can also be confirmed by the embodiment, and the long-term physiological parameter index is provided through the online data processing technology of sequence data learning.
需說明的是，一實施例除了利用上述的矩陣結構與樹狀結構來進行元資料的計算與更新外，更利用資料機率分佈之統計量來做為元資料。It should be noted that, in addition to the above-mentioned matrix structure and tree structure for calculating and updating metadata, an embodiment uses the statistics of the data probability distribution as metadata.
詳言之，本揭露的一實施例所採用的機率分佈為常態分佈，而對應之統計量為平均數與標準差。圖13(a)及圖13(b)是依照本揭露一實施例所繪示的利用數理統計分析方式之統計量記錄元資料，並據以運算生理參數指標的範例。當維度m =2時，本實施例係將所有二維樣本點依數值大小排序整理後，將樣本點的第一維度數值固定於某數值下，利用此時樣本點的第二維度數值所繪製之直方圖如圖13(a)所示。其中，曲線121為常態分佈之函數圖，由圖13(a)可看出，本實施例所採用的常態分佈在第一維度固定下所產生的第二維度資訊，能貼近原始資料的分佈特性。In detail, the probability distribution adopted by an embodiment of the present disclosure is a normal distribution, and the corresponding statistics are an average and a standard deviation. FIG. 13(a) and FIG. 13(b) are diagrams illustrating the use of statistical statistical analysis metadata statistical metadata and an example of calculating physiological parameter indicators according to an embodiment of the present disclosure. When the dimension m =2, in this embodiment, all the two-dimensional sample points are sorted according to the numerical value, and the first dimension value of the sample point is fixed to a certain value, and the second dimension value of the sample point is drawn. The histogram is shown in Figure 13(a). The curve 121 is a function diagram of the normal distribution. As can be seen from FIG. 13( a ), the second dimension information generated by the normal distribution in the first embodiment is close to the distribution characteristics of the original data. .
另一方面，當維度m =3時，將所有三維樣本點之第一維度與第二維度數值固定於某二維組合下，其第三維度資訊之直方圖與對應的常態分佈曲線如圖13(b)所示。其中，從圖13(b)的曲線132同樣可看出利用常態分佈近似資料特性的貼近程度。On the other hand, when the dimension m = 3, the first dimension and the second dimension value of all the three-dimensional sample points are fixed under a certain two-dimensional combination, and the histogram of the third dimension information and the corresponding normal distribution curve are as shown in FIG. (b) is shown. Among them, from the curve 132 of Fig. 13 (b), it can be seen that the normality distribution approximates the closeness of the data characteristics.
在每個視窗利用分佈統計量作為元資料後，可與後續視窗之分佈統計量進行元資料的更新，一實施例所採用之更新公式如下：After each window uses the distribution statistic as the metadata, the metadata of the subsequent window can be updated with the distribution statistics of the subsequent window. The update formula adopted in one embodiment is as follows:
其中，分別代表更新至第t 個視窗時之樣本數與分佈統計量(包括平均數與標準差)；N t +1 ,μ t +1 ,σ t +1 為第t +1個視窗所計算出之樣本數與分佈統計量；則為透過上述兩組元資料更新後，累積至第t +1個視窗之樣本數與分佈統計量。其中，最後記錄的元資料和利用序列式資料學習(sequential data learning)方式所得的最終元資料與利用批次處理方式計算之元資料是完全一致的。among them, Represents the number of samples and distribution statistics (including the average) when updating to the tth window Standard deviation N t +1 , μ t +1 , σ t +1 is the number of samples and distribution statistics calculated by the t +1th window; Then, after updating the above two sets of metadata, the number of samples and the distribution statistics of the t +1th window are accumulated. Among them, the last recorded meta-data and the final metadata obtained by sequential data learning are completely consistent with the meta-data calculated by the batch processing method.
在取得元資料後，本實施例可將該分佈統計量進行標準化處理，利用標準化過之分佈函數在各個區間/區域所佔面積乘上第一維度等於設定值之樣本數量，即能推估在計算多元尺度熵時所需統計的每種樣本點(即向量組合)的出現次數。After obtaining the meta-data, the embodiment can normalize the distribution statistic, and multiply the area occupied by each interval/area by the standardized distribution function by the number of samples whose first dimension is equal to the set value, that is, it can be estimated The number of occurrences of each sample point (ie, vector combination) required to calculate the multivariate entropy.
圖14(a)及圖14(b)是依照本揭露一實施例所繪示的利用分佈統計量運算生理參數指標的時間複雜度分析圖。其中，為了驗証本揭露之利用數理統計分析方式運算生理參數指標方法相較於其他技術的改進，本實施例將原始計算方法、結構化批次計算方法、結構化線上計算方法並僅考量最終反應時間的結構化線上(反應時間)、批次計算分佈統計量之批次估算方法、循序計算分佈統計量並僅考量最終反應時間的線上估算(反應時間)五種方式進行比較。14(a) and 14(b) are time complexity analysis diagrams for calculating a physiological parameter index using a distribution statistic according to an embodiment of the present disclosure. In order to verify the improvement of the method for calculating the physiological parameter index by using the mathematical statistics analysis method of the present disclosure, the original calculation method, the structured batch calculation method, the structured online calculation method and only the final reaction are considered in this embodiment. The timeline (reaction time) of the time, the batch estimation method of the batch calculation distribution statistic, the sequential calculation of the distribution statistic and only the online estimation (reaction time) of the final reaction time are compared in five ways.
由圖14(a)可知，暴力方法所耗用之運算時間呈指數成長，並且甚高於利用元資料進行運算之四種方法；而利用元資料的方法中，又以利用分佈統計量與矩陣結構記錄元資料來運算生理參數指標的方式最有效率。在資料筆數為60,000筆時，原始計算方法需耗時約1,300秒，但從圖14(b)的細部圖示可看出，利用分佈統計量的批次估算方式僅需耗時約110秒，而利用線上估算並僅考量最終反應時間的線上估算(反應時間)也只需85秒，與原始做法相較僅需不到十分之一時間；而矩陣結構的計算方式，以結構化批次計算方法僅需約22秒，結構化線上(反應時間)更進一步縮短至12秒，相較於暴力法此兩方法效率上提升50至100倍。而無論是用分佈統計量估算或是矩陣結構化方式來計算多元尺度熵，採用循序學習的方式相較於批次模式，能更提供生理參數指標。As can be seen from Fig. 14(a), the computation time consumed by the violent method grows exponentially, and is much higher than the four methods of computing using metadata. In the method of using metadata, the distribution statistic and matrix are utilized. It is most efficient to structure the metadata to calculate physiological parameter indicators. When the number of data is 60,000, the original calculation method takes about 1,300 seconds, but as can be seen from the detailed diagram of Figure 14 (b), the batch estimation method using the distribution statistic takes only about 110 seconds. The online estimate (reaction time) that uses online estimates and only considers the final reaction time is only 85 seconds, which is less than one tenth of the time compared to the original practice; and the matrix structure is calculated in a structured batch. The secondary calculation method takes only about 22 seconds, and the structured line (reaction time) is further shortened to 12 seconds, which is 50 to 100 times more efficient than the violent method. Whether using the distribution statistic estimation or the matrix structuring method to calculate the multi-scale entropy, the sequential learning method can provide physiological parameter indicators more than the batch mode.
本揭露可以採線上(on line)或批次(off line)處理。The present disclosure can be processed on an on line or off line.
本揭露之一實施例提出一種內儲程式之電腦可讀取記錄媒體，當電腦載入該程式並執行後，可完成如上所述之方法。One embodiment of the present disclosure provides a computer readable recording medium having a built-in program, and when the computer loads the program and executes it, the method as described above can be completed.
本揭露之一實施例提出一種運算生理參數指標電腦程式產品，當電腦載入該電腦程式並執行後，可完成如上所述之方法。One embodiment of the present disclosure provides a computer program product for computing physiological parameter indicators. When the computer loads the computer program and executes it, the method as described above can be completed.
綜上所述，本揭露之運算生理參數指標的系統、方法、記錄媒體及電腦程式產品可應用於上述長時間生理資訊分析。此方法具有下列特色：In summary, the system, method, recording medium and computer program product for calculating the physiological parameter index of the present disclosure can be applied to the above long-term physiological information analysis. This method has the following characteristics:
1.長時間生理資訊分析：本案針對長時間生理資訊分析中生理參數指標運算效率上提出一改善方法。主要考量在於長時間生理資訊分析相對於短時間生理資訊分析更能反應個案的生理狀態。1. Long-term physiological information analysis: This case proposes an improvement method for the efficiency of physiological parameter index analysis in long-term physiological information analysis. The main consideration is that long-term physiological information analysis is more responsive to the physiological state of the case than short-term physiological information analysis.
2.提高資料儲存的效益：長時間生理參數指標的儲存與計算過程中皆需大量的儲存空間或記憶體空間，本方法原則上僅需利用相對較少的元資料便可計算指標，因此可節省系統的儲存空間；且透過序列式資料學習(sequential data learning)的機制，每個時間區段所需處理的資料量相對較少，加上透過結構化的運算機制能進一步降低運算過程中所需的記憶體空間，而所計算出的元資料會與之前所有時間區段所累計的元資料進行運算更新，因此元資料的量相較於原始資料線性的增長將減緩甚多。2. Improve the efficiency of data storage: a large amount of storage space or memory space is required for the storage and calculation of physiological parameter indicators for a long time. In principle, the method only needs to use relatively small metadata to calculate the index, so Save system storage space; and through the sequential data learning mechanism, the amount of data required to process each time segment is relatively small, and the structure of the operation mechanism can further reduce the operation process. The required memory space, and the calculated metadata will be updated with the metadata accumulated in all previous time segments, so the amount of metadata will be much slower than the linear growth of the original data.
3.提升指標運算速度：本方法以系統化的架構以及序列式資料學習技術解決長時間生理資訊分析所面臨的計算時間冗長問題。藉由系統化的架構記錄元資料，原則上即便是批次處理模式已減少生理參數指標的計算時間；本案更進一步提出利用序列式資料學習技術對每個時間區段的生理資訊進行分析，因為將傳統批次處理的運算時間分散於每個時間區段中，原則上能提供醫療人員病患的長時間生理參數指標。3. Improve the speed of index calculation: This method solves the problem of long calculation time faced by long-term physiological information analysis with systematic architecture and sequential data learning technology. By systematically recording the metadata, in principle, even the batch processing mode has reduced the calculation time of physiological parameter indicators; this case further proposes to use the sequential data learning technology to analyze the physiological information of each time segment because The operation time of the traditional batch processing is dispersed in each time section, and in principle, the long-term physiological parameter index of the medical staff can be provided.
雖然本發明已以實施例揭露如上，然其並非用以限定本發明，任何所屬技術領域中具有通常知識者，在不脫離本發明之精神和範圍內，當可作些許之更動與潤飾，故本發明之保護範圍當視後附之申請專利範圍所界定者為準。Although the present invention has been disclosed in the above embodiments, it is not intended to limit the invention, and any one of ordinary skill in the art can make some modifications and refinements without departing from the spirit and scope of the invention. The scope of the invention is defined by the scope of the appended claims.
20．．．生理資料序列20. . . Physiological data sequence
22．．．長時間生理參數指標twenty two. . . Long-term physiological parameter index
40．．．直方圖40. . . Histogram
60．．．資訊矩陣60. . . Information matrix
62．．．區塊62. . . Block
812．．．儲存媒體812. . . Storage medium
814．．．處理器814. . . processor
816．．．輸入/輸出介面816. . . Input/output interface
818．．．顯示器818. . . monitor
822．．．網路介面822. . . Network interface
824．．．介面824. . . interface
830．．．網路830. . . network
840．．．轉換器840. . . converter
131、132．．．常態分佈曲線131, 132. . . Normal distribution curve
S102~S108．．．本發明一實施例的運算生理參數指標的方法步驟S102~S108. . . Method step of calculating physiological parameter index according to an embodiment of the invention
圖1是依照本發明一實施例所繪示的運算生理參數指標方法的流程圖。FIG. 1 is a flow chart of a method for calculating a physiological parameter index according to an embodiment of the invention.
圖2是依照本發明一實施例所繪示的運算生理參數指標方法的示意圖。FIG. 2 is a schematic diagram of a method for calculating a physiological parameter index according to an embodiment of the invention.
圖3(a)及圖3(b)是依照本發明一實施例所繪示的粗粒化程序的範例。3(a) and 3(b) are diagrams showing an example of a coarse granulation process according to an embodiment of the invention.
圖4是依照本發明一實施例所繪示的計算多元尺度熵之資料結構所對應的直方圖。FIG. 4 is a histogram corresponding to a data structure for calculating multi-scale entropy according to an embodiment of the invention.
圖5是依照本發明一實施例所繪示的使用稀疏矩陣儲存並更新元資料的範例。FIG. 5 illustrates an example of storing and updating metadata using a sparse matrix according to an embodiment of the invention.
圖6是依照本發明一實施例所繪示的利用不斷更新的元資料計算多元尺度熵的示意圖。FIG. 6 is a schematic diagram of calculating multi-scale entropy using continuously updated metadata according to an embodiment of the invention.
圖7(a)、圖7(b)及圖7(c)是依照本發明一實施例所繪示的利用樹狀資料結構記錄元資料，並據以運算生理參數指標的範例。7(a), 7(b), and 7(c) are diagrams illustrating the use of a tree-like data structure to record metadata and calculate physiological parameter indicators according to an embodiment of the invention.
圖8所示為根據一些實施例，適用於執行第1-7圖之方法的運算生理參數指標系統的功能區塊圖。Figure 8 is a functional block diagram of an operational physiological parameter indicator system suitable for performing the methods of Figures 1-7, in accordance with some embodiments.
圖9是依照本發明一實施例所繪示的利用四種評估方式計算多元尺度熵在尺度設定為1的時間分析圖。FIG. 9 is a timing analysis diagram of calculating multi-scale entropy with a scale set to 1 using four evaluation methods according to an embodiment of the invention.
圖10是依照本發明一實施例所繪示的利用結構化方式計算多元尺度熵的比較。FIG. 10 is a comparison of calculating multi-scale entropy using a structured manner according to an embodiment of the invention.
圖11依照本發明一實施例所繪示的利用四種評估方式計算多元尺度熵在尺度設定為1到20所耗用的總時間分析圖。FIG. 11 is a diagram showing a total time analysis diagram for calculating the multi-scale entropy at a scale of 1 to 20 by using four evaluation methods according to an embodiment of the present invention.
圖12是分析圖11中以資料結構化方式計算多元尺度熵的時間。Figure 12 is a graph for analyzing the time at which the multivariate scale entropy is calculated in a data structured manner in Figure 11.
圖13(a)及圖13(b)是依照本發明一實施例所繪示的利用資料分佈之統計量記錄元資料，並據以運算生理參數指標的範例。FIG. 13(a) and FIG. 13(b) are diagrams illustrating the use of statistical data of the data distribution to record metadata and calculate physiological parameter indicators according to an embodiment of the present invention.
圖14(a)及圖14(b)是依照本發明一實施例所繪示的利用五種評估方法計算多元尺度熵在尺度設定為1到20所耗用的總時間分析圖。14(a) and 14(b) are diagrams showing the total time analysis of the multi-scale entropy used to calculate the scale from 1 to 20 using five evaluation methods according to an embodiment of the invention.
S102~S108．．．本發明一實施例之運算生理參數指標的方法步驟S102~S108. . . Method step of calculating physiological parameter index according to an embodiment of the invention
Claims (22)
Priority Applications (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510847888.7A CN105574322B (en) | 2011-06-17 | 2011-12-27 | Physiological parameter index operation system and method |
| CN201110455594.1A CN102831288B (en) | 2011-06-17 | 2011-12-27 | Physiological parameter index operation system and method |
| US13/452,947 US9064039B2 (en) | 2011-06-17 | 2012-04-23 | System, method and recording medium for calculating physiological index |
| US14/710,601 US9996663B2 (en) | 2011-06-17 | 2015-05-13 | System, method and recording medium for calculating physiological index |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US201161497965P | 2011-06-17 | 2011-06-17 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TW201300081A TW201300081A (en) | 2013-01-01 |
| TWI469764B true TWI469764B (en) | 2015-01-21 |
Family
ID=48137217
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW100146537A TWI469764B (en) | 2011-06-17 | 2011-12-15 | System, method, recording medium and computer program product for calculating physiological index |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN105574322B (en) |
| TW (1) | TWI469764B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI800596B (en) * | 2019-01-29 | 2023-05-01 | 宏碁股份有限公司 | Statistic performance evaluation method for different grouping sets |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI489298B (en) * | 2014-01-06 | 2015-06-21 | 國立中央大學 | A multi-scales intrinsic entropies analysis method |
| TWI511072B (en) * | 2014-02-10 | 2015-12-01 | Ind Tech Res Inst | Pathology data processing apparatus and methods |
| CN106343992B (en) * | 2016-08-23 | 2019-02-22 | 清华大学 | Heart rate variance analyzing method, device and purposes |
| CN109199362A (en) * | 2018-08-15 | 2019-01-15 | 哈尔滨理工大学 | The ECG auto-check system and device of cardiac extra systole |
| CN111584063B (en) * | 2019-02-15 | 2023-11-10 | 宏碁股份有限公司 | Method for evaluating statistical efficiency under different grouping sets |
| TWI682354B (en) * | 2019-02-25 | 2020-01-11 | 山衛科技股份有限公司 | Cognitive learning system and method for learning system thinking using the same |
| CN112148763A (en) * | 2019-06-28 | 2020-12-29 | 京东数字科技控股有限公司 | Unsupervised data anomaly detection method and device and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090030762A1 (en) * | 2007-07-26 | 2009-01-29 | Lee Hans C | Method and system for creating a dynamic and automated testing of user response |
| US20090094627A1 (en) * | 2007-10-02 | 2009-04-09 | Lee Hans C | Providing Remote Access to Media, and Reaction and Survey Data From Viewers of the Media |
| US20090271729A1 (en) * | 2006-10-31 | 2009-10-29 | Disetronic Licensing Ag | Method for processing a chronological sequence of measurements of a time dependent parameter |
| US20130127845A1 (en) * | 2010-07-30 | 2013-05-23 | Koninklijke Philips Electronics N.V. | Display and export of individual biplane images |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7882167B2 (en) * | 2005-02-18 | 2011-02-01 | Beth Israel Deaconess Medical Center | Complexity-based dynamical analysis of a network |
| FR2883469B1 (en) * | 2005-03-23 | 2008-02-01 | Eric Georges Lucien Gilli | NECESSARY AND METHOD FOR PROTECTING DRESSINGS AND / OR WOUNDED SKIN AREAS OF THE BODY |
| US7350531B2 (en) * | 2005-07-06 | 2008-04-01 | Thomas Edward Coe | Adjustable crutch |
| US8475387B2 (en) * | 2006-06-20 | 2013-07-02 | Adidas Ag | Automatic and ambulatory monitoring of congestive heart failure patients |
| US8109956B2 (en) * | 2008-03-07 | 2012-02-07 | Medtronic Xomed, Inc. | Systems and methods for surgical removal of tissue |
| EP2276441A1 (en) * | 2008-03-12 | 2011-01-26 | Btampon IP | Web for retention of internal bodily secretions |
-
2011
- 2011-12-15 TW TW100146537A patent/TWI469764B/en active
- 2011-12-27 CN CN201510847888.7A patent/CN105574322B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090271729A1 (en) * | 2006-10-31 | 2009-10-29 | Disetronic Licensing Ag | Method for processing a chronological sequence of measurements of a time dependent parameter |
| US20090030762A1 (en) * | 2007-07-26 | 2009-01-29 | Lee Hans C | Method and system for creating a dynamic and automated testing of user response |
| US20090094627A1 (en) * | 2007-10-02 | 2009-04-09 | Lee Hans C | Providing Remote Access to Media, and Reaction and Survey Data From Viewers of the Media |
| US20130127845A1 (en) * | 2010-07-30 | 2013-05-23 | Koninklijke Philips Electronics N.V. | Display and export of individual biplane images |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI800596B (en) * | 2019-01-29 | 2023-05-01 | 宏碁股份有限公司 | Statistic performance evaluation method for different grouping sets |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105574322B (en) | 2018-07-13 |
| TW201300081A (en) | 2013-01-01 |
| CN105574322A (en) | 2016-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI469764B (en) | System, method, recording medium and computer program product for calculating physiological index | |
| US9996663B2 (en) | System, method and recording medium for calculating physiological index | |
| JP7018133B2 (en) | ECG heartbeat automatic identification classification method based on artificial intelligence | |
| TWI461177B (en) | System, method of detecting abnormal movement of a physical object by periodic signal analysis and computer readable storage medium thereof | |
| JP2022523741A (en) | ECG processing system for depiction and classification | |
| Tutuko et al. | AFibNet: an implementation of atrial fibrillation detection with convolutional neural network | |
| CN113095302B (en) | Depth model for arrhythmia classification, method and device using same | |
| KR20170053693A (en) | Method and apparatus for disease detection | |
| Li et al. | Enabling health monitoring as a service in the cloud | |
| EP2729767A1 (en) | System and method for generating composite measures of variability | |
| CA2862333A1 (en) | Systems and methods for automatic segment selection for multi-dimensional biomedical signals | |
| US10172569B2 (en) | System and method for assisting decisions associated with events relative to withdrawal of life-sustaining therapy using variability measurements | |
| Dalvi et al. | Heartbeat classification system based on neural networks and dimensionality reduction | |
| Wang et al. | Three-heartbeat multilead ECG recognition method for arrhythmia classification | |
| Jacono et al. | Data acquisition and complex systems analysis in critical care: developing the intensive care unit of the future | |
| JP7282161B2 (en) | Advanced cardiac waveform analysis | |
| US20210022620A1 (en) | Systems, Methods and Media for Estimating Compensatory Reserve and Predicting Hemodynamic Decompensation Using Physiological Data | |
| Issa et al. | Heartbeat classification based on single lead-II ECG using deep learning | |
| Singh et al. | Machine learning algorithms for atrioventricular conduction defects prediction using ECG: a comparative study | |
| Shahbudin et al. | Discriminating ECG signals using support vector machines | |
| Cen et al. | Intelligence ECG Monitoring System: Wireless platform and Arrhythmia classification using Residual neural network | |
| Wadhwani et al. | IOT based biomedical wireless sensor networks and machine learning algorithms for detection of diseased conditions | |
| Ardeti et al. | Development of real time ECG monitoring and unsupervised learning classification framework for cardiovascular diagnosis | |
| Vo et al. | Predictive Modelling for Cardiovascular Disease Mortality in Intensive Care Units | |
| Heldt et al. | Model-based data integration in clinical environments |