Functional near infrared spectrum-based depression risk assessment method

Abstract

The invention discloses a functional near infrared spectrum-based depression risk assessment method, which comprises the following steps of: S1, collecting functional near infrared spectrums of depression patients and healthy people, and preprocessing the functional near infrared spectrums to serve as a data set; s2, training a near infrared spectrum extraction model through the data set; and S3, extracting a passing functional near infrared spectrum of a person to be subjected to risk assessment, preprocessing the passing functional near infrared spectrum, inputting the preprocessed passing functional near infrared spectrum into the trained near infrared spectrum extraction model, extracting blood oxygen dynamic characteristics of each region of the brain of the person to be subjected to risk assessment, and performing depression risk assessment. The direct processing of original data is realized, and the capturing capability of the model on the spatial-temporal characteristics of the functional near infrared spectrum signals is improved, so that the depression risk assessment accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of information analysis and processing, and in particular to a method for assessing the risk of depression based on functional near-infrared spectroscopy. Background Art

[0002] In recent years, functional near-infrared spectroscopy (fNIRS) technology has gained widespread attention in the fields of neuroscience and biomedical engineering due to its non-invasiveness, portability and relatively low cost. fNIRS evaluates neural activity by measuring changes in blood oxygen concentration in the cerebral cortex, showing unique advantages. Although fNIRS has great potential in evaluating neural activity, existing methods still face many challenges. Traditional fNIRS analysis methods mainly rely on machine learning models, among which typical representatives include support vector machines (SVM), random forests (RF), and k-nearest neighbors (KNN). These methods usually require manual feature extraction from fNIRS signals to generate low-dimensional data that can be used for classification. Common feature extraction methods include time domain and frequency domain analysis, such as extracting average power, peak amplitude, and time-related changes in blood oxygen concentration. However, the feature extraction process is highly dependent on the experience and domain knowledge of experts. Different experimental designs, data preprocessing methods, etc. may significantly affect the extraction results, which not only increases the complexity of model construction, but also raises the threshold of data analysis, limiting its promotion in non-professional application scenarios.

[0003] In existing studies on fNIRS data processing, there is a significant trade-off between the number of channels and model performance. Although a high number of channels can improve spatial resolution and thus enhance the classification ability of the model, it also significantly increases the cost of hardware equipment and computing resources, especially in practical application scenarios with limited resources, and it is difficult to achieve widespread promotion. In contrast, fNIRS data with low channel numbers often perform poorly in classification accuracy due to the lack of sufficient signal feature extraction, and it is difficult to meet the needs of in-depth analysis of complex brain function activities. Therefore, how to achieve efficient and accurate brain function assessment under low channel number conditions has become a key challenge that needs to be solved urgently. The core issue is whether it is possible to bypass the complicated feature extraction process and directly use the original data for classification and diagnosis, while ensuring the excellent performance of the system in terms of efficiency and robustness. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a depression risk assessment method based on functional near-infrared spectroscopy, designs an end-to-end architecture and spatiotemporal module statistical features, realizes direct processing of raw data, and solves the problem that brain function assessment depends on a high number of channels.

[0005] In order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the present invention is: a method for assessing the risk of depression based on functional near infrared spectroscopy, comprising:

[0006] S1. Collect functional near-infrared spectra of depression patients and healthy people, and use them as data sets after preprocessing;

[0007] S2, training the near infrared spectrum extraction model through the data set;

[0008] S3. Extract the functional near-infrared spectrum of the person to be assessed for risk, input it into the trained near-infrared spectrum extraction model after preprocessing, extract the hemodynamic characteristics of each brain area of ​​the person to be assessed for risk, and conduct depression risk assessment.

[0009] Further: S1 includes:

[0010] S11, collect functional near infrared spectra of depression patients and healthy people during vocabulary fluency test;

[0011] S12, removing motion artifacts in the functional near-infrared spectrum and filtering to obtain a processed functional near-infrared spectrum;

[0012] S13. Convert the processed functional near-infrared spectrum into hemoglobin concentration to obtain low-frequency blood oxygen change information as a data set.

[0013] Further: In S2, the method for processing the input data by the near infrared spectrum extraction model includes:

[0014] S21, inputting the input data into 16 parallel time feature modules respectively to capture key timing features of different frequencies;

[0015] S22, inputting the key time series features of different frequencies into the spatial feature module to identify the blood oxygen dynamics characteristics of each brain region;

[0016] S24, the blood oxygen dynamics characteristics of each brain region are input into the statistical feature module after average pooling to obtain the original depression risk assessment information;

[0017] S25. Expand the original depression risk assessment information into a one-dimensional feature vector, and process it through a fully connected layer with a Softmax activation function to obtain a depression risk assessment result.

[0018] Furthermore: the temporal feature module and the spatial feature module are jointly used as a spatiotemporal feature module, and the spatiotemporal feature module includes a 2D convolution layer, a first batch of normalization layers, a deep 2D convolution layer, a second batch of normalization layers and an ELU activation function layer connected in sequence.

[0019] Further: In S24, the method for processing the hemodynamic characteristics of each brain region by the statistical feature module includes:

[0020] S241, respectively calculating the minimum value, maximum value and weighted average value of the hemodynamic characteristics of each brain region at time t;

[0021] S242, adding the channel signals corresponding to the minimum value, the maximum value and the weighted average value of the hemodynamic characteristics of each brain region at time t element by element to obtain statistical characteristics;

[0022] S243, performing batch normalization, activation function processing, and 2D average pooling on the blood oxygen dynamics characteristics of each brain region in sequence to obtain processed blood oxygen dynamics characteristics;

[0023] S244. Add the statistical features and the processed hemodynamic features element by element to obtain original depression risk assessment information.

[0024] Further: In S241, the calculation expressions of the minimum value, maximum value and weighted average value of the hemodynamic characteristics of each brain region at time t are:

[0025]

[0026] Among them, M min (x i ) represents the minimum value of the hemodynamic characteristics of each brain region at time t, M max (x i ) represents the maximum value of the hemodynamic characteristics of each brain region at time t, It represents the weighted average of the blood oxygen dynamic characteristics of each area of ​​the brain at time t, T represents the time window, sup represents the supremum of the time window T, inf represents the infimum of the time window T, and w(t) represents the weight function.

[0027] The beneficial effects of the present invention are:

[0028] An end-to-end architecture and spatiotemporal module statistical features were designed to enable direct processing of raw data and improve the model's ability to capture the spatiotemporal characteristics of functional near-infrared spectral signals, thereby achieving excellent performance even in the case of low-channel input and improving the accuracy of depression risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flowchart of the depression risk assessment method based on functional near-infrared spectroscopy.

[0030] Figure 2 Schematic diagram of the functional near-infrared spectroscopy collection method.

[0031] Figure 3 Schematic diagram of the near-infrared spectrum extraction model structure.

[0032] Figure 4 Schematic diagram of the statistical feature model structure.

[0033] Figure 5 The figure shows the performance index comparison and index change range of the Depression Dataset under full-channel and low-channel conditions.

[0034] Figure 6 The performance index comparison and index change range diagram of the Stroke Dataset under full-channel and low-channel conditions.

[0035] Figure 7 The following is a graph showing the performance of different models on the Depression Dataset.

[0036] Figure 8 This is a performance comparison chart of the traditional machine learning model when the input is the data after feature extraction and the deep learning model when the input is the original data without feature extraction. DETAILED DESCRIPTION

[0037] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0038] like Figure 1 As shown, in one embodiment of the present invention, a method for assessing depression risk based on functional near infrared spectroscopy is provided, comprising:

[0039] S1. Collect functional near-infrared spectra of depression patients and healthy people, and use them as data sets after preprocessing;

[0040] S2, training the near infrared spectrum extraction model through the data set;

[0041] S3. Extract the functional near-infrared spectrum of the person to be assessed for risk, input it into the trained near-infrared spectrum extraction model after preprocessing, extract the hemodynamic characteristics of each brain area of ​​the person to be assessed for risk, and conduct depression risk assessment.

[0042] In the classification task, when the sample size is relatively small and the number of features is large, the phenomenon of the Curse of Dimensionality is extremely likely to occur, directly affecting the generalization ability of the model and resulting in a significant decline in accuracy. To effectively alleviate this problem, we adopted the feature selection strategy of the interpretive model and combined it with the validation set evaluation mechanism to screen out the optimal channels from the input channels. This method not only improves the generalization performance of the model but also reduces the interference of redundant features on the classification results, ensuring that the model can still maintain a high discriminative ability in the small-sample scenario.

[0043] Specifically, S1 includes:

[0044] S11. Collect the functional near-infrared spectroscopy (FNIRS) of depression patients and healthy individuals during the word fluency test.

[0045] As Figure 2 shown, during the word fluency test, the subjects were required to form words using the Chinese radicals "jia", "bei", and "hua" to stimulate brain activities related to speech generation.

[0046] In this application, a total of 48 probes were used in the FNIRS device. These probes were precisely arranged in the frontal, parietal, and temporal regions of the subjects, covering the brain regions related to functions such as emotional regulation, language processing, and cognitive control. Each probe generated 3-channel data (i.e., HbO, HbR, and HbT), so a total of 144-channel time-series features were extracted, and the sampling frequency was 11 Hz. This frequency is relatively common in FNIRS research, which can effectively balance data accuracy and volume and is suitable for long-term experimental records.

[0047] S12. Remove the motion artifacts in the functional near-infrared spectroscopy and perform filtering to obtain the processed functional near-infrared spectroscopy.

[0048] S13. Convert the processed functional near-infrared spectroscopy into hemoglobin concentration to obtain the low-frequency blood oxygen change information, which is used as the dataset.

[0049] Specifically, in S2, the processing method of the near-infrared spectroscopy extraction model for the input data includes:

[0050] S21. Input the input data into 16 parallel time feature modules respectively to capture the key timing features of different frequencies.

[0051] S22. Input the key timing features of different frequencies into the spatial feature module to identify the blood oxygen dynamics features of each brain region.

[0052] S24. Input the blood oxygen dynamics features of each brain region into the statistical feature module after average pooling to obtain the original depression risk assessment information.

[0053] S25. Expand the original depression risk assessment information into a one-dimensional feature vector, and process it through a fully connected layer with a Softmax activation function to obtain a depression risk assessment result.

[0054] The near-infrared spectral extraction model is a ResNet-like CNN architecture designed for fNIRS-based brain-computer interface (BCI). The model has the following key features: (1) It can be flexibly applied to a variety of BCI paradigms and exhibits strong generalization ability; (2) It can be efficiently trained with only a few channels, significantly reducing the dependence on high-dimensional data; (3) Even when fNIRS data is extremely limited, the near-infrared spectral extraction model can still maintain a high training effect and classification accuracy. The structural diagram of the near-infrared spectral extraction model is shown in the figure. Figure 3 As shown in Figure 1, the design of this architecture not only solves the modeling challenges under low-channel conditions, but also optimizes the sparsity problem of fNIRS signals, so that the model can achieve robust performance on small data sets. The near-infrared spectroscopy extraction model can efficiently capture the key features in fNIRS data while ensuring the compactness and computational efficiency of the model. It is especially suitable for application scenarios where the number of channels is limited but brain activity needs to be accurately identified. The model parameters of the near-infrared spectroscopy extraction model are shown in Table 1:

[0055] Table 1 Near infrared spectrum extraction model parameters (when the training round is 400 rounds)

[0056] Training time 34.73 seconds Number of trainable parameters 5762 Number of non-trainable parameters 80 Total number of parameters 5842

[0057] In particular, the temporal feature module and the spatial feature module are jointly used as a spatiotemporal feature module, and the spatiotemporal feature module includes a 2D convolutional layer, a first batch of normalization layers, a deep 2D convolutional layer, a second batch of normalization layers, and an ELU activation function layer connected in sequence.

[0058] The temporal feature module is used to extract the information features of fNIRS data in the temporal dimension. Considering the lag and slowness of fNIRS data, we set the length of the temporal feature module as short as possible to effectively capture the low-frequency blood oxygen changes of the subjects when performing the VFT task.

[0059] In addition, given the importance of low-frequency information in fNIRS data, we set up 16 “temporal feature modules” to more accurately capture the key timing features of different frequencies.

[0060] The spatial feature module is used to identify the blood oxygen dynamic characteristics of various brain regions and enhance the model's sensitivity to neural activity in specific brain regions, thereby more accurately extracting changes in blood flow and blood oxygen levels in different brain regions and capturing and analyzing regional neural activity.

[0061] like Figure 4 As shown, in S24, the statistical feature module processes the blood oxygen dynamics characteristics of each brain region by:

[0062] S241, respectively calculating the minimum value, maximum value and weighted average value of the hemodynamic characteristics of each brain region at time t;

[0063] The calculation expressions of the minimum, maximum and weighted average values ​​of the hemodynamic characteristics of each brain region at time t are as follows:

[0064]

[0065] Among them, M min (x i ) represents the minimum value of the hemodynamic characteristics of each brain region at time t, M max (x i ) represents the maximum value of the hemodynamic characteristics of each brain region at time t, represents the weighted average of the blood oxygen dynamic characteristics of each brain region at time t, T represents the time window, sup represents the supremum of the time window T, inf represents the infimum of the time window T, and w(t) represents the weight function;

[0066] S242, adding the channel signals corresponding to the minimum value, the maximum value and the weighted average value of the hemodynamic characteristics of each brain region at time t element by element to obtain statistical characteristics;

[0067] S243, performing batch normalization, activation function processing, and 2D average pooling on the blood oxygen dynamics characteristics of each brain region in sequence to obtain processed blood oxygen dynamics characteristics;

[0068] S244. Add the statistical features and the processed hemodynamic features element by element to obtain original depression risk assessment information.

[0069] The statistical feature module not only relies on simple statistics, but also combines functional analysis, integral theory and optimization theory to ensure that the feature extraction of each channel can accurately describe the extreme behavior of the signal and its global trend.

[0070] In one embodiment of the present invention, the dataset DepressionDataset collected using the vocabulary fluency test is shown in Table 2:

[0071] Table 2 DepressionDataset dataset

[0072] male female total Depression patients 2 20 22 Health Personnel 8 9 17 total 10 29 39

[0073] The DepressionDataset dataset contains records of 39 participants, 22 of whom were clinically diagnosed with depression, as shown in the table. The remaining 17 samples were from a healthy control group, and all participants were matched as much as possible in terms of age, gender, education level, etc. to reduce the interference of confounding variables on the results.

[0074] To verify the robustness of the model proposed in this application, we used a public fNIRS dataset from Qingdao University of Science and Technology as a comparison dataset, which contains fNIRS signals of healthy individuals and stroke patients. The data was collected using the NIRSIT portable device, covering 204 probes, each of which recorded data from two hemoglobin channels, HbO and HbR. Compared with our main dataset, this comparison dataset not only has different equipment, but also has a significantly increased number of probes, but only contains two types of blood oxygen signals. This diversity makes it an ideal candidate for testing the adaptability and generalization ability of near-infrared spectral extraction models. In this application, the dataset is named StrokeDataset.

[0075] In this embodiment, we use NirSpark software package to systematically preprocess fNIRS signals. During fNIRS scanning, individuals often have involuntary head movements, and these motion artifacts have a significant impact on functional connectivity analysis.

[0076] After artifact correction, bandpass filtering was further performed, specifically selecting a frequency range of 0.01 to 0.1 Hz to eliminate low-frequency noise caused by physiological activities (such as pulse and respiration). This filtering step is crucial to improving the signal-to-noise ratio of the signal, which can further ensure the accuracy and consistency of the data, thus laying a solid foundation for subsequent analysis.

[0077] After data preprocessing, each experimental sample is converted to a dimension X∈R T ×C, where T = 330 represents the number of sampling time points and C = 144 represents the number of channels. Specifically, each data point x i,j Indicates that at time point t i Next, C j The fNIRS signal intensity of each channel. At present, the main method of using fNIRS for judgment relies on the manual judgment of medical experts, or extracting features on existing data and then applying support vector machine (SVM) for classification. This method relies on the experience and judgment of experts to a certain extent and may be affected by subjective factors. The original data retains the complete information of the original signal without further feature analysis, and directly reflects the spatiotemporal characteristics of the fNIRS signal.

[0078] Based on this, we designed the following experiment. The experimental inputs are shown in Tables 3 and 4:

[0079] Table 3 DepressionDataset Experimental Settings

[0080] full channel Low Channel Feature extraction Near infrared spectrum extraction model √ √ × EEGNet √ √ × KNN √ √ √ MLP √ √ √ RF √ √ √ SVM √ √ √

[0081] Table 4 StrokeDataset Experimental Settings

[0082] full channel Low Channel Feature extraction Near infrared spectrum extraction model √ √ × EEGNet √ √ × KNN √ √ × MLP √ √ × RF √ √ × SVM √ √ ×

[0083] We chose StrokeDataset for comparison, not only to evaluate the performance of the model on different data structures, but also because this dataset only provides two types of features (HbO and HbR), which is different from the feature composition of our main dataset, and has many channels, which can better reflect the adaptability of our near-infrared spectroscopy extraction model under low channels. Through comparative experiments, the results show that the near-infrared spectroscopy extraction model can effectively identify brain functional activities under different probe numbers and feature types, indicating its robustness to noise, probe configuration and signal channels. This further supports the wide applicability and reliability of the near-infrared spectroscopy extraction model in fNIRS brain function research.

[0084] Experimental comparison method: For traditional medical signal processing algorithms, such as support vector machine (SVM), K nearest neighbor (KNN), random forest (RF), and multi-layer perceptron (MLP), we first extract features from the original data. Each sample matrix X is transformed by means of time-frequency domain transformation and statistical feature analysis to generate a new feature matrix X′, which is in the form of:

[0085] X′=f(X)

[0086] Among them, f() represents a specific feature extraction method. These features will be used as input data to evaluate the classification performance of the above classic algorithms after feature extraction.

[0087] Evaluation experiments of various models using the raw data of all channels as input: In order to evaluate the classification ability of various models for fNIRS signals without relying on feature engineering, the raw data without any feature processing was directly input into the near-infrared spectroscopy extraction model, EEGNet, SVM, KNN, RF and MLP models for training and evaluation. The results are shown in Table 5.

[0088] Table 5 Experimental results of each model evaluation using the original data of the full channel as input

[0089]

[0090] Among them, Accuracy_D represents the accuracy of each model for DepressionDataset, and Accuracy_S represents the accuracy of each model for StrokeDataset.

[0091] Experiments to evaluate each model under low-channel raw data input: In practical applications, the number of channels of fNIRS devices is usually limited by hardware resources and experimental environment, so it is important to evaluate the performance of algorithms under low-channel data. In this experiment, we start from the original matrix X∈R T Select CL = 8 channels from ×C to form a new low-channel data set X L ∈R T ×8. In the experimental setting, we directly input the low-channel data into the near-infrared spectrum extraction model, EEGNet, and SVM, KNN, RF, and MLP models for training and testing, keeping the same hyperparameter settings and data preprocessing process as the full-channel experiment. The results are shown in Table 6:

[0092] Table 6 Experimental results of each model evaluation using the original data of the low channel as input

[0093]

[0094] When both machine learning and deep learning models are fed with full-channel raw data: In the full-channel raw data evaluation experiment, we explored the performance of machine learning and deep learning models under the same input conditions under ideal conditions. This experiment assumes that there are sufficient equipment, computing resources, and time conditions. The purpose is to compare the performance of the two types of models under high-dimensional raw data and a sufficient number of probes. Under this condition, the performance of machine learning and deep learning models tends to be consistent, and even in some tasks, machine learning models have a slight advantage. The experimental results show that although machine learning models have a slight advantage in some classification tasks when the amount of computing is sufficient, deep learning models show a wider adaptability without relying on complex preprocessing, which provides an important reference for practical applications.

[0095] When the input is raw data and has 8 channels: The low-channel raw data evaluation experiment focuses on real-world application scenarios, simulating detection and evaluation under conditions where equipment, time, and computing resources are limited. This experiment aims to simulate the actual conditions in which ordinary people cannot rely on high-end equipment and long-term testing in daily health checks. Through low-channel fNIRS data input, this experiment explores how to achieve effective detection and diagnosis while reducing equipment and time costs, thereby providing a cost-effective solution for public health. In this scenario, the performance of traditional machine learning models is greatly reduced due to the limitations of data dimensions and the number of channels (such as Figure 5 (a) with Figure 6(a)), the main indicators such as precision and recall rate have dropped significantly (as shown in Figure 5 (b) with Figure 6 (as shown in (b)).

[0096] in, Figure 5 (a) Performance comparison of different models under full-channel and low-channel conditions on the Depression Dataset; Figure 5 (b) The variation of performance indicators of each model under full channel and low channel input; Figure 6 (a) Performance comparison of different models under full-channel and low-channel conditions on the Stroke Dataset; Figure 6 (b) The change in performance indicators of each model under full-channel and low-channel input.

[0097] However, the near-infrared spectral extraction model can still maintain excellent classification performance by optimizing the input data dimension and channel selection, especially when the channel input is low, such as Figure 7 As shown, Figure 7 (a) is the t-SNE visualization result of the EEGNet model. Figure 7 (b) t-SNE visualization results of the near-infrared spectral extraction model; this shows that the near-infrared spectral extraction model shows extremely strong adaptability and robustness when processing low-channel data, providing a cost-effective solution for the promotion of fNIRS technology in practical applications.

[0098] We use t-SNE to visualize the features extracted by the near-infrared spectroscopy extraction model network, map high-dimensional features to two-dimensional space, and classify and annotate data points according to labels. There is a clear boundary between red and blue data points, which shows that different categories are effectively distinguished in the high-dimensional feature space, proving that the network has strong classification capabilities.

[0099] When the machine learning model input is full-channel feature extraction data and the deep learning model input is 8-channel original data: the experimental results are as follows: Figure 8 As shown in Figure 2, this experiment compares the traditional machine learning model (the feature data carefully extracted under full channels) and the deep learning model (the raw data without processing under low channels). Through this experiment, we have determined the best effect that the traditional machine learning method can achieve after careful preprocessing. The experimental results show that the near-infrared spectroscopy extraction model has good performance in core indicators such as accuracy and recall (such as Figure 8 Both of them surpassed the traditional machine learning models, proving that deep learning models have significant advantages in handling practical tasks and can reduce the complexity of manual feature selection.

[0100] This proves that the proposed near-infrared spectral extraction model can still achieve efficient detection at a low cost under low-channel raw data input, greatly improving its feasibility and universality in practical applications. Through this experiment, we proposed a low-cost and public-friendly detection scheme, which promoted the promotion and application of portable fNIRS detection technology in the field of public health, enabling the general public to afford the detection costs of brain dysfunction, especially in the screening of mental illnesses such as depression. The low-cost and high-efficiency characteristics of the near-infrared spectral extraction model meet the needs of large-scale applications, greatly reducing the technical threshold, so that groups with limited economic conditions can enjoy such detection services, thereby expanding the coverage of mental health screening. With the increasing severity of global mental health problems, especially among adolescents and high-pressure working groups, early detection of depression has become particularly important. The near-infrared spectral extraction model provides a practical solution for large-scale health screening, especially for non-professional medical environments such as schools and community health centers. By simplifying the detection process, the near-infrared spectral extraction model is expected to become a convenient and efficient mental health screening tool, helping to detect potential problems early and intervene in time, thereby reducing the burden on the medical system and improving the mental health level of the group.

[0101] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for assessing depression risk based on functional near infrared spectroscopy, characterized in that: include: S1. Collect functional near-infrared spectra of depression patients and healthy people, and use them as data sets after preprocessing; S2, training the near infrared spectrum extraction model through the data set; S3. Extract the functional near-infrared spectrum of the person to be assessed for risk, input it into the trained near-infrared spectrum extraction model after preprocessing, extract the hemodynamic characteristics of each brain area of ​​the person to be assessed for risk, and conduct depression risk assessment.

2. The method for assessing depression risk based on functional near infrared spectroscopy according to claim 1, characterized in that: S1 includes: S11, collect functional near infrared spectra of depression patients and healthy people during vocabulary fluency test; S12, removing motion artifacts in the functional near-infrared spectrum and filtering to obtain a processed functional near-infrared spectrum; S13. Convert the processed functional near-infrared spectrum into hemoglobin concentration to obtain low-frequency blood oxygen change information as a data set.

3. The method for assessing depression risk based on functional near infrared spectroscopy according to claim 1, characterized in that: In S2, the near infrared spectrum extraction model processes the input data in the following ways: S21, inputting the input data into 16 parallel time feature modules respectively to capture key timing features of different frequencies; S22, inputting the key time series features of different frequencies into the spatial feature module to identify the blood oxygen dynamics characteristics of each brain region; S24, the blood oxygen dynamics characteristics of each brain region are input into the statistical feature module after average pooling to obtain the original depression risk assessment information; S25. Expand the original depression risk assessment information into a one-dimensional feature vector, and process it through a fully connected layer with a Softmax activation function to obtain a depression risk assessment result.

4. The method for assessing depression risk based on functional near infrared spectroscopy according to claim 3, characterized in that: The temporal feature module and the spatial feature module are jointly used as the spatiotemporal feature module, which includes a 2D convolutional layer, a first batch of normalization layers, a deep 2D convolutional layer, a second batch of normalization layers, and an ELU activation function layer connected in sequence.

5. The method for assessing depression risk based on functional near infrared spectroscopy according to claim 3, characterized in that: In S24, the statistical feature module processes the blood oxygen dynamics characteristics of each brain region by: S241, respectively calculating the minimum value, maximum value and weighted average value of the hemodynamic characteristics of each brain region at time t; S242, adding the channel signals corresponding to the minimum value, the maximum value and the weighted average value of the hemodynamic characteristics of each brain region at time t element by element to obtain statistical characteristics; S243, performing batch normalization, activation function processing, and 2D average pooling on the blood oxygen dynamics characteristics of each brain region in sequence to obtain processed blood oxygen dynamics characteristics; S244. Add the statistical features and the processed hemodynamic features element by element to obtain original depression risk assessment information.

6. The method for assessing depression risk based on functional near infrared spectroscopy according to claim 5, characterized in that: In S241, the calculation expressions for the minimum value, maximum value and weighted average value of the hemodynamic characteristics of each brain region at time t are: Among them, M min (x i ) represents the minimum value of the hemodynamic characteristics of each brain region at time t, M max (x i ) represents the maximum value of the hemodynamic characteristics of each brain region at time t, It represents the weighted average of the blood oxygen dynamic characteristics of each area of ​​the brain at time t, T represents the time window, sup represents the supremum of the time window T, inf represents the infimum of the time window T, and w(t) represents the weight function.

Images