Efficacy prediction model construction, prediction method, device and electronic equipment

Abstract

The present invention discloses the construction of a therapeutic effect prediction model, a prediction method, a device and an electronic device, wherein the method for constructing the therapeutic effect prediction model includes: obtaining the EEG data of the sample object in the awake state and the sleep state during the interictal period; determining the nonlinear dynamic characteristics of the sample object according to the EEG data of the sample object; processing the nonlinear dynamic characteristics of the sample object to determine the feature vector of the sample object, wherein the feature vector of the sample object is composed of the nonlinear dynamic characteristics after screening; training a preset therapeutic effect prediction model according to the feature vector and the feature vector of the sample object, and determining the therapeutic effect prediction model when the prediction accuracy reaches the preset standard. The constructed therapeutic effect prediction model can be used to determine whether the object to be predicted is suitable for receiving a specific clinical treatment, and patients who are not suitable for surgery can avoid unnecessary expenses.

Description

Technical Field

[0001] The present invention relates to the field of big data analysis technology, and in particular to the construction of therapeutic effect prediction models, prediction methods, devices and electronic equipment. Background Art

[0002] Epilepsy is a clinical syndrome caused by abnormal discharges of highly synchronized brain neurons due to various reasons. There are currently about 65 million epilepsy patients worldwide, and about 9 million epilepsy patients in my country. There are about 650,000 to 700,000 new cases of epilepsy each year. Epilepsy can seriously affect the physical and mental health of patients and bring inconvenience to their lives. Epilepsy patients can use one or more drugs to control epileptic seizures. About 30% of patients still have poor results after long-term drug treatment and recurrent seizures, developing into drug-resistant epilepsy. The treatment of refractory epilepsy mainly includes traditional surgical treatment, ketogenic diet and neuroregulation. Vagus nerve stimulation (VNS) is a neuroregulatory therapy for refractory epilepsy that has been clinically proven to be safe and long-term effective. However, since the therapeutic mechanism of VNS is still unclear, its individual differences in efficacy are very large and have a high degree of uncertainty. Existing statistical analysis results show that about 5%-9% of patients with drug-resistant epilepsy have their seizures completely controlled after VNS treatment, but about 10% of patients have no effect after VNS treatment, and the remaining patients show varying degrees of reduction in seizure frequency. Specifically, 50%-60% of patients with drug-resistant epilepsy who receive VNS treatment can achieve a 50% reduction in seizure frequency. If the uncertainty and individual differences in the efficacy of VNS in the treatment of drug-resistant epilepsy can be solved, there will be a certain prediction of its treatment effect before VNS surgery, which is of great significance to the patient's rehabilitation, disease cognition, and the conservation of medical resources. Researchers have been trying to screen out patients suitable for VNS surgery through preoperative evaluation.

[0003] Currently, there is no clear method for screening patients for VNS surgery indications for clinical use. Studies on factors related to VNS efficacy based on electroencephalography (EEG), electrocardiography (ECG), magnetic resonance imaging (MRI), patient demographic characteristics (gender, age, etc.), clinical history (disease course), and seizure characteristics (including seizure type, seizure frequency, lesion location, etc.) have yielded inconsistent conclusions.

[0004] As the main tool for recording brain waves, electroencephalograph plays a fundamental role in epilepsy research. EEG signals are also the gold standard for clinical diagnosis. Compared with imaging examinations, they are more convenient and simple to operate, and the cost is not high. Therefore, the processing of EEG signals is widely used in predicting the efficacy of VNS in treating refractory epilepsy. The current idea of ​​using EEG signals to predict the efficacy of patients is: for the enrolled patients, the long-term EEG signals of the patients are collected, and the long-term efficacy of the patients after VNS treatment is combined. The characteristics of the patients' preoperative EEG signals are analyzed, and the characteristics with statistical differences between the EEG signals of patients with different efficacy groups are found as sensitive factors for screening patients for VNS surgery or predicting the efficacy.

[0005] Existing methods based on EEG signal processing mainly focus on analyzing the characteristic results of the signal in the time domain, frequency domain, time-frequency analysis, connectivity, etc., and then statistically analyzing the various characteristics of the preoperative EEG signals of patients with different efficacy classifications. Statistical test methods are usually used to analyze the relationship between a single variable and the treatment response. However, simple time-frequency analysis, connectivity and other features often cannot express the complex characteristics of the dynamic changes of the nervous system, nor can they fully reflect the potential activities of the nervous system. At the same time, they are easily affected by the patient's nervous system activities and events. At the same time, the statistical test method itself may ignore the comprehensive effects of different features, and the research conclusions on the same problem are inconsistent and cannot be repeated. In summary, the research using the above methods has not been applied clinically. Summary of the invention

[0006] In order to solve the technical problems existing in the prior art, the present invention provides the construction of an efficacy prediction model, a prediction method, a device and an electronic device, which can accurately and efficiently screen patients with drug-refractory epilepsy suitable for VNS surgery, and help patients who are not suitable for VNS avoid the risks of long-term stimulation and high surgical costs.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a method for constructing a therapeutic efficacy prediction model, the method comprising the following steps:

[0009] Obtaining EEG data of the sample subjects during the awake and sleeping states;

[0010] Determining the nonlinear dynamic characteristics of the sample object according to the EEG data of the sample object;

[0011] Processing the nonlinear dynamic characteristics of the sample object to determine a feature vector of the sample object, wherein the feature vector of the sample object is composed of the screened nonlinear dynamic characteristics;

[0012] The preset efficacy prediction model is trained according to the feature vectors of the sample objects, and when the prediction accuracy reaches a preset standard, the efficacy prediction model is determined.

[0013] In a possible implementation, the nonlinear dynamics feature includes a complexity feature, an entropy feature, and a fractal analysis feature, and determining the multiple nonlinear dynamics features of the sample object according to the EEG data of the sample object includes:

[0014] Performing window segmentation on the EEG data of the sample object to determine EEG data window segments;

[0015] The nonlinear dynamics characteristic complexity, entropy and fractal analysis are calculated for the computer data window fragments respectively to determine the complexity characteristic, entropy characteristic and fractal analysis characteristic of the sample object.

[0016] In a possible implementation, the processing of the nonlinear dynamic characteristics of the sample object to determine the characteristic vector of the sample object includes:

[0017] According to the efficacy of the specific clinical treatment of the sample objects, the sample objects are divided into positive samples and negative samples, wherein the positive samples are sample objects whose efficacy meets the preset conditions, and the negative samples are sample objects whose efficacy does not meet the preset conditions;

[0018] Performing a difference significance analysis on the nonlinear dynamic characteristics to determine the nonlinear dynamic characteristics with significant differences between the positive sample and the negative sample;

[0019] The recursive feature elimination feature selection method is adopted to screen the nonlinear dynamic features with significant differences according to the importance of the efficacy of specific clinical treatments corresponding to the nonlinear dynamic features with significant differences, and the nonlinear dynamic features after screening constitute the feature vectors of the sample objects.

[0020] In a possible implementation, the recursive feature elimination feature selection method is used to screen the nonlinear dynamic features with significant differences according to the importance of the efficacy of the specific clinical treatment corresponding to each nonlinear dynamic feature, and the nonlinear dynamic features after screening constitute the feature vector of the sample object, specifically including:

[0021] Inputting nonlinear dynamic features with significant differences into the classification model to obtain weight coefficients corresponding to each nonlinear dynamic feature, wherein the weight coefficients correspond to the importance of the nonlinear dynamic feature to the efficacy of a specific clinical treatment;

[0022] The weight coefficients are sorted, and the nonlinear dynamic features corresponding to the minimum weight coefficients are removed. The nonlinear dynamic features after screening constitute the feature vector of the sample object.

[0023] In a possible implementation, before determining the nonlinear dynamic characteristics of the sample object according to the EEG data of the sample object, the method further includes:

[0024] The EEG data of the sample subject is subjected to filtering, noise reduction, re-reference and segment screening processing.

[0025] In a second aspect, the present invention provides a method for predicting therapeutic efficacy, the method comprising the following steps:

[0026] Obtain the feature vector of the object to be predicted;

[0027] The feature vector of the object to be predicted is input into the efficacy prediction model constructed by the method described in any embodiment of the first aspect to obtain the efficacy prediction result of the object to be predicted, so as to determine whether the object to be predicted is suitable for receiving specific clinical treatment.

[0028] In a third aspect, the present invention provides a device for constructing a therapeutic efficacy prediction model, the device comprising:

[0029] The first acquisition module is used to acquire the EEG data of the sample subject during the awake state and the sleeping state;

[0030] A feature module, used for determining the nonlinear dynamic features of the sample object according to the EEG data of the sample object;

[0031] A processing module, used for processing the nonlinear dynamic characteristics of the sample object to determine a feature vector of the sample object, wherein the feature vector of the sample object is composed of the screened nonlinear dynamic characteristics;

[0032] The training module is used to train a preset efficacy prediction model according to the feature vector of the sample object, and determine the efficacy prediction model when the prediction accuracy reaches a preset standard.

[0033] In a possible implementation, the feature module is specifically used to:

[0034] Performing window segmentation on the EEG data of the sample object to determine EEG data window segments;

[0035] The nonlinear dynamics characteristic complexity, entropy and fractal analysis are calculated for the computer data window fragments to determine the complexity characteristic, entropy characteristic and fractal analysis characteristic of the sample object.

[0036] In a possible implementation, the processing module is specifically configured to:

[0037] According to the efficacy of the specific clinical treatment of the sample objects, the sample objects are divided into positive samples and negative samples, wherein the positive samples are sample objects whose efficacy meets the preset conditions, and the negative samples are sample objects whose efficacy does not meet the preset conditions;

[0038] Performing a difference significance analysis on the nonlinear dynamic characteristics to determine the nonlinear dynamic characteristics with significant differences between the positive sample and the negative sample;

[0039] A recursive feature elimination feature selection method is adopted to screen the nonlinear dynamic features with significant differences for a preset number of times according to the importance of the efficacy of specific clinical treatments corresponding to the nonlinear dynamic features with significant differences, and the nonlinear dynamic features after screening constitute the feature vectors of the sample objects.

[0040] In a fourth aspect, the present invention provides a therapeutic effect prediction device, the device comprising:

[0041] The second acquisition module is used to obtain the feature vector of the object to be predicted;

[0042] A prediction module is used to input the feature vector of the object to be predicted into the efficacy prediction model constructed by any device embodiment of the third aspect to obtain the efficacy prediction result of the object to be predicted, so as to determine whether the object to be predicted is suitable for receiving specific clinical treatment.

[0043] In a fifth aspect, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0044] Memory, used to store computer programs;

[0045] A processor, configured to implement the steps of the method for constructing a therapeutic efficacy prediction model according to any one of the embodiments of the first aspect when executing a program stored in the memory;

[0046] Alternatively, when used to execute a program stored in the memory, the steps of the therapeutic effect prediction method as described in the second aspect are implemented.

[0047] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for predicting therapeutic effects according to any one of the embodiments of the first aspect are implemented;

[0048] Alternatively, when executed by a processor, the steps of the therapeutic efficacy prediction method as described in the second aspect are implemented.

[0049] The above technical solution provided by the embodiment of the present application has the following advantages compared with the prior art:

[0050] The embodiment of the present application provides a method for constructing a therapeutic effect prediction model, which obtains the EEG data of the sample object during the waking state and the sleeping state; determines the nonlinear dynamic characteristics of the sample object based on the EEG data of the sample object; processes the nonlinear dynamic characteristics of the sample object to determine the feature vector of the sample object, and the feature vector of the sample object is composed of the screened nonlinear dynamic characteristics; trains the preset therapeutic effect prediction model based on the feature vector of the sample object, and determines the therapeutic effect prediction model when the prediction accuracy reaches the preset standard. The constructed therapeutic effect prediction model can be used to determine whether the object to be predicted is suitable for receiving a specific clinical treatment, and patients who are not suitable for surgery can avoid unnecessary expenses. For example, for patients with drug-resistant epilepsy, patients with drug-resistant epilepsy suitable for VNS surgery can be accurately and efficiently screened, helping patients who are not suitable for VNS to avoid the risk of long-term stimulation and high surgical costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of a simple flow chart of a therapeutic effect prediction model and prediction method provided by the present invention;

[0052] Figure 2 A schematic diagram of a method for constructing a therapeutic efficacy prediction model provided by the present invention;

[0053] Figure 3 A schematic diagram of another therapeutic effect prediction construction and prediction method provided by the present invention;

[0054] Figure 4 Schematic diagram of the process for obtaining complexity features;

[0055] Figure 5 Schematic diagram of the process for obtaining entropy features;

[0056] Figure 6 To obtain the schematic diagram of the fractal analysis process;

[0057] Figure 7 Schematic diagram of the process for determining the eigenvector;

[0058] Figure 8 A schematic diagram of another process for determining a characteristic vector;

[0059] Fig. 9 A schematic flow chart of a method for predicting therapeutic efficacy provided by the present invention;

[0060] Fig.10 A schematic diagram of the structure of a device for constructing a therapeutic effect prediction model provided by an embodiment of the present invention;

[0061] Fig.11 A schematic diagram of the structure of another device for constructing a therapeutic efficacy prediction model provided by an embodiment of the present invention;

[0062] Fig.12 A schematic diagram of the structure of a therapeutic effect prediction device provided by an embodiment of the present invention;

[0063] Fig.13 A schematic diagram of the structure of an electronic device is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] To facilitate understanding of the embodiments of the present invention, specific embodiments will be further explained below in conjunction with the accompanying drawings. The embodiments do not constitute a limitation on the embodiments of the present invention.

[0066] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of more restrictions, the elements defined by the sentence "comprise one..." do not exclude the existence of other identical elements in the process, method, article or equipment including the elements. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0067] Existing methods based on EEG signal processing mainly conduct statistical analysis on the preoperative time domain, frequency domain, connectivity and other characteristics of patients with different efficacy classifications, and usually use statistical test methods to analyze the relationship between a single variable and the treatment response. On the one hand, because the information of the EEG signals reflected by these characteristics is relatively messy and relatively simple, it is easy to be interfered by the patient's own EEG activity, and it is difficult to express the potential information of the EEG signals. At the same time, a single feature is used to analyze the relationship with the efficacy, which may ignore the comprehensive effect of different features. In this way, the research conclusions on the same problem are often inconsistent and cannot be repeated. In addition, the various indicators of EEG signals are affected by the sleep and wakefulness states, and the changes are very obvious. Comprehensive analysis of multiple states will better predict the results. In summary, the research using the above methods has not been applied clinically.

[0068] In response to the technical problems mentioned in the background technology, the present invention proposes a method based on EEG nonlinear dynamic characteristics and machine learning, which analyzes the changes in the dynamic characteristics of the brain, a multi-dimensional complex dynamic system, through nonlinear dynamics, and captures the complexity of the brain's dynamic changes. This can measure the information capacity of signal fragments and reflect the potential activity characteristics of neurons. And through the nonlinear dynamic characteristics of the whole brain and local brain areas in different states, combined with the machine learning system, the accuracy of efficacy prediction is improved.

[0069] To achieve the above objectives, the present invention uses the nonlinear dynamic characteristics of EEG data in different states (sleep and wakefulness) to construct a machine learning classification model to predict the efficacy of VNS patients. Figure 1 As shown, first obtain the training sample (the EEG data of the sample object), and build a binary classification model based on the extracted nonlinear dynamic features; then obtain the prediction sample (the EEG data of the object to be predicted), and then extract its nonlinear dynamic features and input them into the constructed binary classification model to determine whether the object to be predicted is valid. The detailed process is introduced below:

[0070] Specifically, the present invention provides a method for constructing a therapeutic effect prediction model. Figure 2 As shown, Figure 2 A method for constructing a therapeutic efficacy prediction model provided by an embodiment of the present invention is as follows: Figure 2 As shown, the method comprises the following steps:

[0071] Step 210, obtaining the EEG data of the sample subject during the waking state and the sleeping state.

[0072] Patients who receive specific clinical treatments (such as VNS) are referred to as sample subjects. Each index of the EEG signal is affected by the sleep and wakefulness states, and the changes are very obvious. In order to ensure the accuracy of subsequent predictions, the EEG data of the interictal period in each preset time period in both the wakefulness and sleep states are obtained from the long-term EEG signals collected from the patient before surgery. For example, the EEG data of the interictal period of at least two hours in both the wakefulness and sleep states are collected. That is, the EEG data of the sample subject can be obtained from the preoperative EEG data of the sample subject. Among them, the preoperative EEG data of the sample subject can be the EEG data obtained by using clinical EEG equipment before the sample subject receives specific clinical treatment.

[0073] In order to facilitate the processing of EEG data, it is necessary to convert the format of EEG data. Figure 3 Another method for predicting therapeutic efficacy provided by an embodiment of the present invention is as follows: Figure 3 As shown, in order to increase the quality of EEG data and make subsequent predictions more accurate, it is necessary to preprocess the EEG data after format conversion, and finally form the preprocessed EEG data into complete EEG data segments. Specifically, the preprocessing includes: filtering and denoising, re-referencing and segment screening of the EEG data of the sample object.

[0074] The following is a brief introduction to the preprocessing process of filtering noise reduction, re-reference and clip screening:

[0075] Filtering can remove data outside the preset frequency and leave data within the preset range. After filtering, baseline drift, power frequency interference and high-frequency noise in EEG data can be removed. For example, you can choose bandpass filtering and set the filtering parameters in advance, such as 2Hz~30Hz. After filtering, data with a frequency less than 2Hz and a frequency greater than 30Hz will be removed, and the remaining data will have a frequency between 2Hz and 30Hz.

[0076] The acquired EEG data contains multiple signal components from independent signal sources mixed in the same signal. Therefore, the EEG signal of each channel may include signal components generated by electrooculographic activity, head movement, and electromyographic activity in addition to the signal components generated by EEG activity. In order to make the signal components generated by EEG activity occupy a larger proportion in the EEG data of each channel, the signal components generated by non-EEG activity are removed, that is, the signal components generated by electrooculographic activity, head movement, and electromyographic activity mentioned above are removed.

[0077] Re-referencing means resetting the reference electrode, which can usually be set to a nose tip electrode, a top electrode, etc. The EEG signal of each channel can be the potential difference between the active electrode corresponding to the channel and the reference electrode (such as a top electrode). In this case, for two active electrodes that are closer and farther away from the reference electrode, when the potential changes at the two active electrodes caused by EEG activity are the same, the potential difference between the active electrode farther away from the reference electrode and the reference electrode will naturally be greater than the potential difference between the active electrode closer to the reference electrode and the reference electrode. This difference is caused by the selection of the reference electrode, has nothing to do with EEG activity, and causes certain errors in the EEG data. Therefore, the reference electrode can be reselected through re-referencing processing, thereby reducing the error of the EEG data caused by the selection of the reference electrode.

[0078] In order to improve the accuracy of data processing, a preset time length can be set, the EEG data can be segmented according to the preset time length, and problematic data segments can be removed, and the screened data segments can be formed into complete EEG data segments as preprocessed EEG data.

[0079] Step 220: determining the nonlinear dynamic characteristics of the sample object based on the EEG data of the sample object.

[0080] The nonlinear dynamic characteristics include complexity characteristics, entropy characteristics and fractal analysis characteristics. The EEG data of the sample object is window-segmented to determine the computer data window segments. Then, the nonlinear dynamic characteristics complexity, entropy and fractal analysis are calculated for the obtained computer data window segments to determine the complexity characteristics, entropy characteristics and fractal analysis characteristics of the sample object.

[0081] In nonlinear dynamic characteristics, complexity measures information capacity by extracting potential patterns from signals, while entropy measures the characteristic distribution of signals based on Shannon information theory. Both can quantitatively describe the activity of the complex brain system and effectively reflect the potential activity characteristics of the nervous system. Fractal analysis can show the degree of chaos in the brain system and also reflect the predictable time scale of the chaotic system. Compared with healthy people, the neuronal activity of epileptic patients has more uncertainty and randomness, and VNS can promote the state changes of the nervous system of effective subjects, so nonlinear dynamic characteristics can well reflect some state changes of the nervous system that are not easy to observe.

[0082] For some nonlinear dynamic features involving frequency domain signals (such as wavelet entropy and Hilbert-Huang entropy), the features can be divided into different frequency bands to analyze the nonlinear dynamics of different frequency bands. In addition, for all the nonlinear dynamic features obtained, their whole-brain mean results can be calculated, or they can be focused on local channels (brain regions) to calculate the characteristics of nonlinear dynamics in each brain region.

[0083] The following is a brief introduction to complexity features, entropy features and fractal analysis features:

[0084] The definitions of complexity features, entropy features and fractal analysis features are detailed in Tables 1, 2 and 3 below, respectively.

[0085] (1) Complexity characteristics:

[0086] Table 1 Complexity characteristics

[0087]

[0088] LZ mean is the mean complexity, LZC media is the median complexity, LZC mid-point is the median complexity, LZC k-mean The complexity of k-means is the four coarse-grained methods. By comparing the signal amplitude and the threshold, the original signal is converted into a 0-1 sequence. The first three complexities only require a threshold T. d (LZC mean The threshold is the mean, LZC media The threshold is the median of the signal sequence, LZC mid-point The threshold is the mean of the maximum and minimum values), while LZC k-mean Two thresholds are required (by setting two centroids z(1) and z(2) and calculating the distance between the data point and the centroid to determine the position of the data point).

[0089] Figure 4 This is a flow chart of obtaining complexity features. It should be noted that the specific window segmentation in step 410 is different depending on the complexity features obtained. Figure 4 As shown in the figure, after obtaining the coarse-grained new sequence, the complexity c(n) of the new sequence is calculated, which represents the number of different patterns in the original sequence {x(n)}. The above four types of complexity can be represented by c(n):

[0090]

[0091] Among them, LZC includes LZC mean , LZC media For LZC mid-point and LZC k-mean .

[0092] PLZC (sort complexity) does not consider the absolute amplitude value but the order of the values. The sorting method has the advantages of simple calculation, strong robustness and low complexity. The sorting complexity characteristics are obtained by the following methods:

[0093] (a) The sorting process is to determine the number of combination points m, and then establish m! possible permutation patterns. After matching the original signal with the permutation pattern, it can be converted into a finite sequence of symbols, whose possible elements are 1 to m!.

[0094] (b) Initialize the metric value of PLZC. Calculate the complexity of the new sequence c(n), which represents the number of different patterns in the original digital sequence {x(n)}. PLZC can be represented by c(n):

[0095]

[0096] (2) Entropy characteristics:

[0097] Table 2 Entropy characteristics

[0098]

[0099]

[0100] Figure 5 This is a flow chart of obtaining entropy features. It should be noted that different entropy features are obtained, and the specific window segmentation is also different. Figure 5 As shown:

[0101] Shannon entropy (ShEn) can measure the probability density based on the probability distribution of amplitude values. Its calculation formula is:

[0102]

[0103] Among them, p(s i ) represents the random variable S equal to s i The probability of , i represents the data point, and a represents the logarithmic base.

[0104] like Figure 5 As shown in Figure 1, ApEn (approximate entropy) can describe the unpredictability of the signal. By reconstructing the EEG data signal into a phase space with an embedding dimension of m, when m increases to m+1, the growth rate of the number of phase space modes within a predetermined value r is estimated. The calculation formula is:

[0105]

[0106]

[0107] Denotes the embedding vector X m (j) and X m(i) The probability that the distance between them is less than r. The ApEn calculation is easily affected by the data length N, the similarity capacity r and the embedding dimension m. Generally, in EEG signal processing, n is 1000, r is 0.15-0.25 of the signal standard deviation, and m is 2-3. It can also be called the total number of template matches.

[0108] SampEn (sample entropy) is developed from approximate entropy. It also reconstructs the signal into phase space and calculates the growth rate of the number of phase space modes as the embedding dimension increases. Its calculation formula is:

[0109]

[0110]

[0111]

[0112] Sample entropy is similar to approximate entropy, and the parameters are defined the same as approximate entropy. Denotes the embedding vector X m (j) and X m (i) The probability that the distance between them is less than r, where N is the data length. However, the sample entropy removes the self-matching in the data and keeps the total number of patterns the same when embedding dimension m and m+1, which can be understood as:

[0113] When the embedding dimension is m+1, the total number of template matches is A m (r); When the embedding dimension is m, the total number of template matches is B m (r).

[0114] PE (sort entropy) is a quantitative complexity algorithm that can reflect the local structure of dynamic time series. It has the characteristics of simplicity, strong noise resistance, and low computational complexity. First, the EEG data signal needs to be reconstructed, {X(i+(m-1)τ)}, τ represents the time delay, and m represents the embedding dimension; then the sequence is sorted in ascending order and mapped to one of the m! sortings. Its calculation formula is:

[0115]

[0116]

[0117] where p j Represents the probability of occurrence of different sequence combinations, n j Indicates the number of times the j-th sort occurs.

[0118] The above are all based on the entropy analysis of time domain signals, such as Figure 5As shown in the figure, WE (wavelet entropy) and HHSE (Hilbert-Huang Spectrum Entropy) are entropy value analyses based on signal time-frequency analysis. WE (wavelet entropy) can distinguish different specific states of itself. When the EEG signal has stronger regularity, that is, the neuronal activity is synchronized, the wavelet entropy value will decrease. First, the wavelet energy of each EEG signal scale j is calculated. Divide each wavelet energy by the total energy to obtain the relative wavelet energy of each scale j:

[0119]

[0120] The calculation formula is:

[0121]

[0122] in, Represents the relative wavelet energy of each scale j. The value of WE depends on the wavelet basis function, the number of decomposition layers n and the length of the data N. The most important one is the wavelet basis function. In actual use, the wavelet basis function needs to be reasonably selected according to the patient's EEG data.

[0123] HHSE (Hilbert-Huang Spectral Entropy) is a spectral entropy analysis combined with Hilbert transform, which uses the empirical mode decomposition method to process nonlinear and non-stationary signals. For the EEG signal x(t), it can be expressed as:

[0124]

[0125] Among them, imf represents a series of intrinsic functions of the signal, and then the imf component is Hilbert transformed to obtain:

[0126] Z(t)=imf(t)+iH[imf(t)]=a(t)e i∫ω(t)dt ,

[0127] in,

[0128] h(ω)=∫H(ω,t)dt,

[0129] ω(t) and a(t) represent the instantaneous frequency and amplitude of IMF, respectively. represents the normalized Hilbert spectrum, and then the Shannon entropy algorithm is applied to the Hilbert spectrum to obtain the Hilbert Huang spectrum entropy:

[0130]

[0131] Hilbert-spectral entropy can provide higher resolution in both time domain and frequency domain to capture frequency changes in the signal. And because of its decomposition mode, it will not produce oscillation mode distortion due to instantaneous amplitude, and can better show the potential information of EEG signals.

[0132] (3) Fractal analysis characteristics:

[0133] Table 3 Fractal analysis characteristics

[0134]

[0135] Figure 6 In order to obtain a schematic diagram of the fractal analysis process, it should be noted that the specific window segmentation is different depending on the fractal analysis features obtained. Figure 6 As shown:

[0136] RQA (recursive quantitative analysis) can intuitively reflect the motion laws and trends of dynamic systems and give accurate quantification of these characteristics. First, the EEG signal is reconstructed in phase space to obtain an m-dimensional embedding vector {X(n+(m-1)τ)}, where τ is the time delay, and then a recursive graph is constructed to form a matrix R ij :

[0137] R ij =θ(ε-r ij )

[0138] Where ε is the distance threshold and ij is the time. Finally, the recursive graph quantization parameters are calculated, such as the recursive rate RR, the deterministic DET, etc.

[0139]

[0140]

[0141] The Lyaponuv index, whose numerical value represents the degree of system mixing, also requires reconstructing the EEG data into phase space, {X(t+(m-1)τ)}, and then representing the system in the form of n first-order differential equations:

[0142]

[0143] Select two trajectories with similar starting points from the system and calculate the divergence of their adjacent trajectories. The calculation formula is:

[0144]

[0145]

[0146] DF is the Jacobian matrix of each point, ω(x0,t)=x(x0+Δx0,t)-x(x0,t).

[0147] The correlation dimension is mainly based on the GP algorithm to calculate the correlation, which also requires reconstructing the EEG signal data into the phase space, {X(i+(m-1)τ)}, and then estimating the value from the experimental time series by observing how the number of points in a sphere with a radius of ε in space changes as the radius decreases to zero.

[0148]

[0149]

[0150] Where θ is the step function, r is the distance, and D2 is the correlation dimension.

[0151] Detrended fluctuation analysis can detect the physical characteristics of time series. First, a new series is established based on the original EEG signal data. The new sequence is divided into non-overlapping equal-length subintervals of length s, and then a polynomial regression fit is performed on the data of each subinterval v to obtain the local trend function y v (i) Eliminate the trend in each sub-interval and calculate its variance mean to obtain the q-order fluctuation function F of the entire sequence q (s):

[0152]

[0153] Finally, determine the scaling exponent h of the wave function q , that is, there is a power law relationship: h q is the Hurst index, when h q =0.5, the sequence is an independent process; when 0.5≤h q <1, the sequence has long-range correlation; when h q When <0.5, the series has negative long-range correlation, that is, anti-persistence.

[0154] Step 230, processing the nonlinear dynamic characteristics of the sample object to determine a feature vector of the sample object, wherein the feature vector of the sample object is composed of the screened nonlinear dynamic characteristics.

[0155] like Figure 7 As shown, nonlinear dynamic feature selection and feature vector determination include the following steps:

[0156] Step 710, according to the efficacy of specific clinical treatment of the sample objects, the sample objects are divided into positive samples and negative samples, wherein the positive samples are sample objects whose efficacy meets the preset conditions, and the negative samples are sample objects whose efficacy does not meet the preset conditions.

[0157] According to the clinical treatment effect of VNS, the characteristic data of the sample objects are divided into an effective group (epileptic seizure control rate is greater than or equal to the preset threshold) and an ineffective group (e.g., epilepsy control rate is less than the preset threshold). For example, the effective group corresponds to a reduction of ≥50% in epilepsy seizures after VNS surgery, and the ineffective group corresponds to a reduction of <50% in epilepsy seizures after VNS surgery. The effective group is also called a positive sample, and the ineffective group is called a negative sample.

[0158] Step 720 , performing a difference significance analysis on the nonlinear dynamic features to determine the nonlinear dynamic features with significant differences between the positive sample group and the negative sample group.

[0159] An appropriate statistical test method (such as the Mann-Whitney U test) is used to perform a significant difference analysis of all nonlinear dynamic features between the positive sample and the negative sample groups, and nonlinear dynamic features with statistically significant differences are retained.

[0160] In addition, in order to avoid the influence of trained weights due to different orders of magnitude or dimensions, the nonlinear dynamic characteristics are normalized.

[0161] Step 730, using a recursive feature elimination feature selection method, multiple nonlinear dynamic features with significant differences are screened according to the importance of the efficacy of specific clinical treatments corresponding to the nonlinear dynamic features with significant differences, and the screened nonlinear dynamic features constitute the feature vector of the sample object.

[0162] Specifically, Figure 8 As shown in Figure 2, the nonlinear dynamic characteristics with significant differences are determined by the following method to determine the feature vector of the sample object:

[0163] Inputting nonlinear dynamic features with significant differences into the classification model to obtain weight coefficients corresponding to each nonlinear dynamic feature, wherein the weight coefficients correspond to the importance of the nonlinear dynamic feature to the efficacy of a specific clinical treatment;

[0164] The weight coefficients are sorted, and the nonlinear dynamic features corresponding to the minimum weight coefficients are removed. The nonlinear dynamic features after screening constitute the feature vector of the sample object.

[0165] That is, the recursive feature elimination and feature selection method is used to calculate the importance scores of VNS therapeutic effects corresponding to nonlinear dynamic features with statistically significant differences, and different nonlinear dynamic features are ranked according to their importance to VNS therapeutic effects, and the features with high importance are selected to construct the feature vector of the sample.

[0166] Step 240, training a preset efficacy prediction model based on the feature vector of the sample object, and determining the efficacy prediction model when the prediction accuracy reaches a preset standard.

[0167] Specifically, the sample objects are marked to obtain labels of the sample objects, and a preset efficacy prediction model is trained according to the feature vectors of the sample objects and the corresponding labels of the sample objects. When the prediction accuracy reaches a preset standard, the efficacy prediction model is determined.

[0168] A method for constructing a therapeutic efficacy prediction model provided by an embodiment of the present invention obtains EEG data of the sample subject during the waking state and the sleeping state during the interictal period; determines the nonlinear dynamic characteristics of the sample object based on the EEG data of the sample object; processes the nonlinear dynamic characteristics of the sample object to determine the feature vector of the sample object, and the feature vector of the sample object is composed of the screened nonlinear dynamic characteristics; trains a preset therapeutic efficacy prediction model based on the feature vector of the sample object, and determines the therapeutic efficacy prediction model when the prediction accuracy reaches the preset standard. The constructed therapeutic efficacy prediction model can be used to determine whether the object to be predicted is suitable for receiving a specific clinical treatment, and patients who are not suitable for surgery can avoid unnecessary expenses. For example, for patients with drug-resistant epilepsy, patients with drug-resistant epilepsy who are suitable for VNS surgery can be accurately and efficiently screened, helping patients who are not suitable for VNS to avoid the risk of long-term stimulation and high surgical costs.

[0169] The above introduces the method for constructing the efficacy prediction model. The following introduces the specific process of prediction using the prediction model constructed by the above method.

[0170] For more details on the efficacy prediction method, see Figure 3 The process shown, specifically, the steps of the efficacy prediction method include:

[0171] Step 310: Obtain a feature vector of the object to be predicted.

[0172] The patient to be predicted is called the object to be predicted. The method of obtaining the feature vector of the object to be predicted is as described in steps 210-240. First, the same EEG data collection requirements are performed on the patient to be predicted, and the same data processing and EEG nonlinear dynamic feature extraction are performed. The feature vector of the patient to be predicted is constructed according to the feature vector construction method of the prediction model.

[0173] Step 320, input the feature vector of the subject to be predicted into the efficacy prediction model to obtain the efficacy prediction result of the subject to be predicted, so as to determine whether the subject to be predicted is suitable for receiving a specific clinical treatment.

[0174] The efficacy prediction model, also known as a binary classification prediction model, inputs the feature vector of the object to be predicted in step 310 into the efficacy prediction model in step 240, and uses the binary classification prediction model of the efficacy of VNS in treating epilepsy to predict the patient data samples to be predicted, thereby obtaining the results of the VNS in treating epilepsy efficacy prediction.

[0175] An embodiment of the present invention provides a method for predicting therapeutic effects, which obtains a feature vector of an object to be predicted; inputs the feature vector of the object to be predicted into a therapeutic effect prediction model to obtain a therapeutic effect prediction result of the object to be predicted, so as to determine whether the object to be predicted is suitable for receiving a specific clinical treatment. In this way, patients who are not suitable for surgery can avoid unnecessary expenses. For example, for patients with drug-resistant epilepsy, patients with drug-resistant epilepsy who are suitable for VNS surgery can be accurately and efficiently screened, helping patients who are not suitable for VNS to avoid the risk of long-term stimulation and high surgical costs.

[0176] The above are examples of the construction of the efficacy prediction model and the prediction method provided in this application. The following is an introduction to other examples of the construction of the efficacy prediction model and prediction provided in this application. Please refer to the following for details.

[0177] Fig.10 A schematic diagram of a device for constructing a therapeutic efficacy prediction model provided by an embodiment of the present invention is shown in FIG. Fig.10 As shown, the therapeutic effect prediction device includes: a first acquisition module 101, a feature module 102, a processing module 103 and a training module 104.

[0178] The first acquisition module 101 is used to acquire the EEG data of the sample subject during the waking state and the sleeping state.

[0179] The feature module 102 is used to determine the nonlinear dynamic features of the sample object according to the EEG data of the sample object.

[0180] The processing module 103 is used to process the nonlinear dynamic characteristics of the sample object and determine the characteristic vector of the sample object. The characteristic vector of the sample object is composed of the screened nonlinear dynamic characteristics.

[0181] The training module 104 is used to train a preset efficacy prediction model according to the feature vectors of the sample objects, and determine the efficacy prediction model when the prediction accuracy reaches a preset standard.

[0182] 11 In one example, the feature module 102 is specifically used for:

[0183] Performing window segmentation on the EEG data of the sample object to determine the EEG data window segments;

[0184] The nonlinear dynamics feature complexity, entropy and fractal analysis are calculated for the computer data window fragments to determine the complexity feature, entropy feature and fractal analysis feature of the sample object.

[0185] In another example, the processing module 103 is specifically configured to:

[0186] According to the efficacy of specific clinical treatment of the sample objects, the sample objects are divided into positive samples and negative samples, wherein the positive samples are sample objects whose efficacy meets the preset conditions, and the negative samples are sample objects whose efficacy does not meet the preset conditions;

[0187] Perform a significant difference analysis on the nonlinear dynamic features to determine the nonlinear dynamic features with significant differences between the positive sample and the negative sample groups;

[0188] The recursive feature elimination feature selection method is adopted to screen multiple nonlinear dynamic features with significant differences for a preset number of times according to the importance of the efficacy of specific clinical treatments corresponding to the nonlinear dynamic features with significant differences. The screened nonlinear dynamic features constitute the feature vector of the sample object.

[0189] Fig.11 A schematic diagram of another therapeutic effect prediction device provided by an embodiment of the present invention is shown in FIG. Fig.11 As shown, the therapeutic effect prediction device also includes: a preprocessing module 905.

[0190] The preprocessing module 905 is used to perform filtering, noise reduction, re-referencing and segment screening on the EEG data of the sample object before determining the nonlinear dynamic characteristics of the sample object based on the EEG data of the sample object.

[0191] The functions performed by each component in the therapeutic effect prediction model construction device provided in the embodiment of the present invention have been described in detail in any of the above method embodiments, and therefore will not be repeated here.

[0192] An embodiment of the present invention provides a device for constructing a therapeutic efficacy prediction model, which obtains EEG data of a sample subject during the waking state and the sleeping state during the interictal period; determines the nonlinear dynamic characteristics of the sample subject based on the EEG data of the sample subject; processes the nonlinear dynamic characteristics of the sample subject to determine the feature vector of the sample subject, wherein the feature vector of the sample subject is composed of the screened nonlinear dynamic characteristics; trains a preset therapeutic efficacy prediction model based on the feature vector of the sample subject and the label of the corresponding sample subject, and determines the therapeutic efficacy prediction model when the prediction accuracy reaches the preset standard. The constructed therapeutic efficacy prediction model can be used to determine whether the object to be predicted is suitable for receiving a specific clinical treatment, and patients who are not suitable for surgery can avoid unnecessary costs. For example, for patients with drug-resistant epilepsy, patients with drug-resistant epilepsy who are suitable for VNS surgery can be accurately and efficiently screened, helping patients who are not suitable for VNS to avoid the risk of long-term stimulation and high surgical costs.

[0193] In addition, the present application also provides a therapeutic effect prediction device, which uses the therapeutic effect prediction model constructed by the therapeutic effect prediction model construction device introduced above to make predictions, specifically, Fig.12As shown, including:

[0194] The second acquisition module 906 is used to acquire a feature vector of the object to be predicted.

[0195] The prediction module 907 is used to input the feature vector of the object to be predicted into the efficacy prediction model to obtain the efficacy prediction result of the object to be predicted, so as to determine whether the object to be predicted is suitable for receiving a specific clinical treatment.

[0196] The functions performed by each component in the therapeutic effect prediction device provided by the embodiment of the present invention have been described in detail in any of the above method embodiments, and will not be repeated here.

[0197] An embodiment of the present invention provides a therapeutic effect prediction device, which obtains a feature vector of an object to be predicted; inputs the feature vector of the object to be predicted into a therapeutic effect prediction model to obtain a therapeutic effect prediction result of the object to be predicted, so as to determine whether the object to be predicted is suitable for receiving a specific clinical treatment. In this way, patients who are not suitable for surgery can avoid unnecessary expenses. For example, for patients with drug-resistant epilepsy, patients with drug-resistant epilepsy who are suitable for VNS surgery can be accurately and efficiently screened, helping patients who are not suitable for VNS to avoid the risk of long-term stimulation and high surgical costs.

[0198] like Fig.13 As shown, an embodiment of the present application provides an electronic device, including a processor 131 , a communication interface 132 , a memory 133 and a communication bus 134 , wherein the processor 131 , the communication interface 132 , and the memory 133 communicate with each other through the communication bus 134 .

[0199] A memory 133, for storing computer programs;

[0200] In one embodiment of the present application, the processor 131 is used to implement the steps of the efficacy prediction method provided by any of the aforementioned method embodiments when executing the program stored in the memory 133.

[0201] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the efficacy prediction method provided in any of the aforementioned method embodiments are implemented.

[0202] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0203] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0204] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A method for constructing a therapeutic effect prediction model, characterized in that: The method comprises the following steps: Obtaining EEG data of the sample subjects during the awake and sleeping states; Determining the nonlinear dynamic characteristics of the sample object according to the EEG data of the sample object; Processing the nonlinear dynamic characteristics of the sample object to determine a feature vector of the sample object, wherein the feature vector of the sample object is composed of the screened nonlinear dynamic characteristics; According to the feature vector of the sample object, a preset efficacy prediction model is trained, and when the prediction accuracy reaches a preset standard, the efficacy prediction model is determined; the preset efficacy prediction model is a binary classification model; The nonlinear dynamic characteristics include complexity characteristics, entropy characteristics and fractal analysis characteristics. The multiple nonlinear dynamic characteristics of the sample object are determined according to the EEG data of the sample object, including: Performing window segmentation on the EEG data of the sample object to determine EEG data window segments; Calculating the complexity, entropy and fractal analysis of nonlinear dynamic characteristics of the EEG data window segments respectively to determine the complexity characteristics, entropy characteristics and fractal analysis characteristics of the sample object; The complexity features include: mean complexity, median complexity, median complexity, k-means complexity and sorting complexity; the complexity features are used to extract the potential patterns of the sample objects and measure the information capacity; the entropy features include: Shannon entropy, approximate entropy, sample entropy, sorting entropy, wavelet entropy and Hilbert-Huang spectrum entropy; the entropy features are used to measure the characteristic distribution of the sample objects; the fractal analysis features include: recursive quantitative analysis, Lyaponuv index, correlation dimension and detrended fluctuation analysis; the fractal analysis features are used to express the degree of chaos of the brain system and reflect the time scale of prediction; The processing of the nonlinear dynamic characteristics of the sample object to determine the characteristic vector of the sample object includes: According to the efficacy of the specific clinical treatment of the sample objects, the sample objects are divided into positive samples and negative samples, wherein the positive samples are sample objects whose efficacy meets the preset conditions, and the negative samples are sample objects whose efficacy does not meet the preset conditions; Performing a difference significance analysis on the nonlinear dynamic characteristics of the sample objects to determine the nonlinear dynamic characteristics with significant differences between the positive sample group and the negative sample group; The recursive feature elimination feature selection method is adopted to screen the nonlinear dynamic features with significant differences according to the importance of the efficacy of specific clinical treatments corresponding to the nonlinear dynamic features with significant differences, and the nonlinear dynamic features after screening constitute the feature vectors of the sample objects.

2. The method according to claim 1, characterized in that The recursive feature elimination feature selection method is used to screen the nonlinear dynamic features with significant differences according to the importance of the efficacy of the specific clinical treatment corresponding to each nonlinear dynamic feature. The nonlinear dynamic features after screening constitute the feature vector of the sample object, which specifically includes: Inputting nonlinear dynamic features with significant differences into the classification model to obtain weight coefficients corresponding to each nonlinear dynamic feature, wherein the weight coefficients correspond to the importance of the nonlinear dynamic feature to the efficacy of a specific clinical treatment; The weight coefficients are sorted, and the nonlinear dynamic features corresponding to the minimum weight coefficients are removed. The nonlinear dynamic features after screening constitute the feature vector of the sample object.

3. The method according to claim 1, characterized in that Before determining the nonlinear dynamic characteristics of the sample object according to the EEG data of the sample object, the method further includes: The EEG data of the sample subject is subjected to filtering, noise reduction, re-reference and segment screening processing.

4. A method for predicting therapeutic efficacy, characterized in that: The method comprises the following steps: Obtain the feature vector of the object to be predicted; The feature vector of the object to be predicted is input into the efficacy prediction model constructed by the method according to any one of claims 1 to 3 to obtain the efficacy prediction result of the object to be predicted, so as to determine whether the object to be predicted is suitable for receiving specific clinical treatment.

5. A device for constructing a therapeutic effect prediction model, characterized in that: The device comprises: The first acquisition module is used to acquire the EEG data of the sample subject during the awake state and the sleeping state; A feature module, used to determine the nonlinear dynamic characteristics of the sample object according to the EEG data of the sample object; the nonlinear dynamic characteristics include complexity characteristics, entropy characteristics and fractal analysis characteristics; A processing module, used for processing the nonlinear dynamic characteristics of the sample object to determine a feature vector of the sample object, wherein the feature vector of the sample object is composed of the screened nonlinear dynamic characteristics; A training module, used to train a preset efficacy prediction model according to the feature vector of the sample object, and determine the efficacy prediction model when the prediction accuracy reaches a preset standard; the preset efficacy prediction model is a binary classification model; The feature module is specifically used for: Performing window segmentation on the EEG data of the sample object to determine EEG data window segments; Calculating the complexity, entropy and fractal analysis of nonlinear dynamic characteristics of the EEG data window segments respectively to determine the complexity characteristics, entropy characteristics and fractal analysis characteristics of the sample object; The complexity features include: mean complexity, median complexity, median complexity, k-means complexity and sorting complexity; the complexity features are used to extract the potential patterns of the sample objects and measure the information capacity; the entropy features include: Shannon entropy, approximate entropy, sample entropy, sorting entropy, wavelet entropy and Hilbert-Huang spectrum entropy; the entropy features are used to measure the characteristic distribution of the sample objects; the fractal analysis features include: recursive quantitative analysis, Lyaponuv index, correlation dimension and detrended fluctuation analysis; the fractal analysis features are used to express the degree of chaos of the brain system and reflect the time scale of prediction; The processing module is specifically used for: According to the efficacy of the specific clinical treatment of the sample objects, the sample objects are divided into positive samples and negative samples, wherein the positive samples are sample objects whose efficacy meets the preset conditions, and the negative samples are sample objects whose efficacy does not meet the preset conditions; Performing a difference significance analysis on the nonlinear dynamic characteristics to determine the nonlinear dynamic characteristics with significant differences between the positive sample and the negative sample; A recursive feature elimination feature selection method is adopted to screen the nonlinear dynamic features with significant differences for a preset number of times according to the importance of the efficacy of specific clinical treatments corresponding to the nonlinear dynamic features with significant differences, and the nonlinear dynamic features after screening constitute the feature vectors of the sample objects.

6. A therapeutic effect prediction device, characterized in that: The device comprises: The second acquisition module is used to acquire the feature vector of the object to be predicted; A prediction module is used to input the feature vector of the object to be predicted into the efficacy prediction model constructed by the device as described in claim 5 to obtain the efficacy prediction result of the object to be predicted, so as to determine whether the object to be predicted is suitable for receiving specific clinical treatment.

7. An electronic device, characterized in that: The electronic device carries the resource scheduling system, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, configured to implement the steps of the method for constructing a therapeutic efficacy prediction model according to any one of claims 1 to 3 when executing the program stored in the memory; Alternatively, when used to execute a program stored in a memory, the steps of the therapeutic effect prediction method as described in claim 4 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a therapeutic efficacy prediction model according to any one of claims 1 to 3 are implemented; Alternatively, when executed by a processor, the steps of the therapeutic effect prediction method as claimed in claim 4 are implemented.

Images