Method for identifying multiple categories of mental disorders

Abstract

The invention relates to a method for identifying multiple categories of mental disorders, and aims to realize accurate classification and rapid diagnosis of mental disorders through collection and analysis of multi-modal data. The method comprises the following steps of collecting multi-modal data of a patient, wherein the multi-modal data comprises physiological data, behavior data, medical image data and psychological questionnaire data; filtering, de-noising, standardizing and other pre-processing are carried out on the collected data, key features of each mode are extracted, and feature vectors are formed; generating a uniform feature vector by weighting and fusing the feature vectors of different modals; and training and classifying the feature vectors based on a deep learning model (MLP), and outputting the confidence of each mental disorder category. And finally, generating a diagnosis report according to a classification result, prompting high-risk and medium-risk categories, and providing treatment suggestions. According to the method, the limitation of a single data mode can be overcome, multi-dimensional information is comprehensively integrated, and the accuracy and efficiency of diagnosis are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of health monitoring, and in particular to a method for identifying multiple categories of mental disorders. Background Art

[0002] Mental disorders are a type of disease that seriously affects human mental health and social functions, including depression, anxiety, schizophrenia and other types. Early screening, accurate diagnosis and classification of these diseases are of great significance to the patient's rehabilitation and treatment plan formulation. At present, the identification methods of mental disorders mainly rely on the following technical means:

[0003] (1) Psychological questionnaire assessment: Psychological questionnaires (such as PHQ-9, GAD-7, PANSS, etc.) are the most commonly used tools for assessing mental disorders. A series of questions are used to understand the patient's psychological state, and the scores are compared with the diagnostic criteria to draw preliminary conclusions.

[0004] (2) Physiological data analysis: including electroencephalogram (EEG), heart rate variability (HRV), skin conductance, etc., which have gradually been used in the research and auxiliary diagnosis of mental disorders.

[0005] (3) Medical imaging analysis: Imaging technologies such as magnetic resonance imaging (MRI) and functional magnetic resonance imaging (fMRI) can be used to analyze abnormalities in brain structure and function and are often used to study diseases such as schizophrenia and major depression.

[0006] (4) Behavioral data analysis: By collecting behavioral data such as patients’ facial expressions, voice signals, and movement trajectories, we study their relationship with mental disorders.

[0007] Existing technologies still have the following deficiencies in the identification and diagnosis of mental disorders:

[0008] Limitations of single data modality: Most current technologies only utilize single modality data (such as psychological questionnaires, EEG signals, or imaging data), which has limited information and is difficult to fully reflect the complex pathology of mental disorders.

[0009] The correlation between data from different modalities is not effectively utilized, resulting in inaccurate or incomplete diagnostic results.

[0010] Insufficient integration of subjective and objective information: Subjective assessment methods such as psychological questionnaires lack objective support from physiological and behavioral data and cannot effectively identify disguised or hidden symptoms.

[0011] Although physiological data and imaging data are objective, their correspondence with actual symptoms is complex, and their diagnostic efficacy is insufficient when used alone.

[0012] Low diagnostic efficiency: Although methods such as medical image analysis are reliable, they are costly and time-consuming, making them difficult to adapt to the needs of rapid screening.

[0013] Traditional methods require doctors to make manual judgments based on multi-source data, which is time-consuming and easily affected by personal experience and bias.

[0014] Existing deep learning models are not widely used:

[0015] At present, some studies have attempted to use deep learning methods to classify single-modal data, but the deep fusion technology of multi-modal data is not yet mature.

[0016] Existing models have poor adaptability to small sample data, resulting in insufficient generalization ability in practical applications. Summary of the invention

[0017] In one aspect, a method for identifying multiple categories of mental disorders comprises the following steps:

[0018] Multimodal data collection: Collect multimodal data of patients, including physiological data (EEG signals, EEG (t), heart rate X HR (t), skin conductance X EDA (t)), behavioral data (facial expression X FACE 、Speech signal X AUDIO (t)), medical imaging data X MRI , and psychological questionnaire data X Q = {Q 1 ,XQ 2 ,…,Q n};

[0019] Data preprocessing: filtering, denoising, and standardizing the collected multimodal data to form clean data;

[0020] Feature extraction: Extract key features from each modality data to form EEG feature vectors F EEG , heart rate feature vector F HR 、Image feature vector F M RI and questionnaire feature vector F Q ;

[0021] Feature fusion: The feature vectors of all modalities are weightedly fused to generate a unified feature vector V;

[0022] Classification model training and application: Train and classify unified feature vectors based on deep learning models, and output the confidence level of each mental disorder category;

[0023] Result interpretation: A diagnostic report is generated based on the classification results, indicating high-risk and medium-risk categories and providing treatment recommendations.

[0024] On the other hand, the data collection specifically includes:

[0025] Physiological data collection: EEG signals X EEG (t): sampling frequency f s =500Hz, where t represents time and the signal unit is microvolt (μV);

[0026] Heart RateX HR (t): sampling frequency f s =1Hz, unit is beats per minute (BPM);

[0027] Skin ConductanceX EDA (t): The unit is Siemens (μS), the sampling resolution is 10 -3 μS;

[0028] Behavioral data collection: Facial expression sequence X FACE : Obtained through video, frame rate f video =30FPS; voice signal X AUDIO (t): sampling frequency f s =16kHz, unit is amplitude value;

[0029] Imaging data acquisition: Magnetic resonance imaging dataX MRI , resolution 1×1×1mm 3 ;

[0030] Psychological questionnaire data: questionnaire score vector X Q = {q 1 ,q 2 ,…,q n}, where q i ∈[0,4] represents the score of the i-th item, and n is the total number of questions in the questionnaire.

[0031] On the other hand, the data preprocessing includes: EEG signal filtering: EEG (t) Use a bandpass filter H(f) with a filtering frequency range of [0.5,50] Hz:

[0032]

[0033] in, and Respectively represent Fourier transform and inverse transform, f represents frequency; Y EEG (t) Filtered EEG signal;

[0034] Speech signal preprocessing: AUDIO (t) Extract Mel frequency cepstrum coefficients:

[0035]

[0036] Among them, C k : kth order cepstrum coefficient; X n : filter energy value; N: number of filters.

[0037] Data standardization: For all modal data X = {x 1 ,x 2 ,…,x m}Perform zero mean standardization:

[0038]

[0039] Among them, x i is the original data, μ is the mean, σ is the standard deviation, Z i is the standardized data.

[0040] On the other hand, the feature extraction includes:

[0041] EEG signal frequency domain features F EEG The steps include: calculating the power spectrum density of the EEG signal and extracting the power of the δ, θ, α, and β bands:

[0042]

[0043] Among them, P δ ,P θ ,P α ,P β are the power values ​​of each band, |X(f)| 2 is the power spectral density of the signal; X(f) is the spectrum of the EEG signal; f is the frequency;

[0044] EEG signal frequency domain features F EEG =[P δ ,P θ ,P α ,P β ];

[0045] Image feature F MRI The following steps are included: Segment the image into k brain regions R 1 ,R 2 ,…,R k ;

[0046] Extract brain region volumes:

[0047] Where V i : the volume of the ith brain region; ρ(x,y,z): the pixel value of each pixel.

[0048] Image feature F MRI =[V 1 ,V 2 ,…,V i ];

[0049] Extract the heart rate feature vector F HR Includes the following steps: Average heart rate: where μ HR : average heart rate; T: duration of signal acquisition; X HR (t): heart rate signal;

[0050] Heart Rate Variability:

[0051]

[0052] Among them, HRV LF ,HRV HF : Heart rate variability low-frequency and high-frequency power;

[0053] X HR (f): spectrum of heart rate signal;

[0054] Heart rate feature vector F HR =[μ HR ,HRV LF ,HRV HF ];

[0055] Extract the questionnaire feature vector F Q The steps are as follows: Calculate the total score:

[0056]

[0057] Where, S: total score of the questionnaire; W: i : The weight of the i-th item; Q i : The score of the ith item.

[0058] Calculate sub-score: S depression =∑ i∈D W i Q i ,S anxiety =∑ i∈A W i Q i , where S depression ,S anxiety : Depression and anxiety scores; D, A: A collection of questions related to depression and anxiety;

[0059] Questionnaire feature vector F Q =[S,S depression ,S anxiety ].

[0060] On the other hand, the feature vectors of all modes are weighted and fused to generate a unified feature vector V:

[0061] Feature fusion formula:

[0062] Among them, W EEG 、w HR 、w MRI 、w Q is the weight coefficient of each mode, satisfying ∑w i =1.

[0063] On the other hand, the classification model uses a deep learning structure and is trained based on the following cross entropy loss function:

[0064]

[0065] Where m is the number of samples, K is the number of mental disorder categories, and y ik is the true category label, is the k-th class probability predicted by the model, θ is the model parameter, and L(θ) is the loss function.

[0066] On the other hand, by introducing the regularization term to optimize the model:

[0067] L′(θ)=L(θ)+λ||θ|| 2 ,

[0068] Among them, λ is the regularization coefficient and ||θ|| is the L2 norm of the parameter.

[0069] On the other hand, the classification model is a multi-layer perceptron, including: input layer: input feature vector v; hidden layer: using ReLU activation function:

[0070] h (l) =max(0,W (l) j (l-1) +b (l) ),

[0071] Among them, W (l) is the weight matrix, b (l) is bias;

[0072] Output layer: Output category probability through Softmax function:

[0073]

[0074] On the other hand, the confidence is calculated based on the probability output by the model:

[0075]

[0076] Among them, C k is the confidence of the kth category, when C k ≥75%, marked as high risk category; when 50%≤C k <75%, marked as medium risk category.

[0077] On the other hand, the method is suitable for hospital outpatient screening, mental health assessment equipment and telemedicine platforms for rapid identification of multiple categories of mental disorders such as depression, anxiety and schizophrenia.

[0078] Beneficial effects:

[0079] The present invention collects the patient's physiological data (electroencephalogram, heart rate, skin conductance), behavioral data (facial expression, voice signal), medical imaging data (MRI) and psychological questionnaire data (PHQ-9, GAD-7, etc.), comprehensively covers the multidimensional characteristics of mental disorders, and combines the weighted fusion technology of multimodal data to avoid the limitations of existing single-modal data methods, greatly improving the comprehensiveness and scientificity of diagnosis. At the same time, the combination of subjective and objective information effectively reduces the risk of misdiagnosis and missed diagnosis.

[0080] The present invention uses a deep learning model (MLP) to achieve automatic identification of multiple categories of mental disorders through feature extraction, cross entropy loss function training and Softmax classification, reducing the bias of manual judgment and greatly improving the diagnostic efficiency. At the same time, the L2 regularization term optimization model is introduced to enhance the generalization ability, so that it can maintain good classification performance under small sample data, which is suitable for actual clinical and telemedicine scenarios.

[0081] The present invention can be widely used in hospital outpatient screening, mental health assessment equipment and telemedicine platforms, and is suitable for rapid screening and accurate diagnosis of multiple mental disorders such as depression, anxiety, schizophrenia, etc. Especially in the identification of mild or early mental disorders, the present invention can achieve early screening, early diagnosis, and early intervention with comprehensive data fusion and efficient algorithm analysis, helping patients get timely treatment and improve their quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0083] Figure 1 A flowchart for a method of identifying multiple categories of mental disorders;

[0084] Figure 2 The type of data processed in the data preprocessing step. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0086] The following is combined with Figure 1 and Figure 2 , the specific implementation methods of the present invention are described in detail.

[0087] A method for identifying multiple categories of mental disorders, comprising the following steps:

[0088] 1. Multimodal data collection: Collect multimodal data of patients, including physiological data (EEG signal X EEG (t), heart rate X HR (t), skin conductance X EDA (t)), behavioral data (facial expression X FACE 、Speech signal X AUDIO (t)), medical imaging data X MRI , and psychological questionnaire data X Q = {Q 1 ,Q 2 ,…,Q n};

[0089] Data collection specifically includes:

[0090] Physiological data collection: EEG signals X EEG (t): sampling frequency f s =500Hz, where t represents time and the signal unit is microvolt (μV);

[0091] Heart RateX HR (t): sampling frequency f s =1Hz, unit is beats per minute (BPM);

[0092] Skin ConductanceX EDA (t): The unit is Siemens (μS), the sampling resolution is 10 -3 μS;

[0093] Behavioral data collection: Facial expression sequence X FACE : Obtained through video, frame rate f video =30FPS; voice signal XAUDIO (t): sampling frequency f s =16kHz, unit is amplitude value;

[0094] Imaging data acquisition: Magnetic resonance imaging dataX MRI , resolution 1×1×1mm 3 ;

[0095] Psychological questionnaire data: questionnaire score vector X Q = {q 1 ,q 2 ,…,q n}, where q i ∈[0,4] represents the score of the i-th item, and n is the total number of questions in the questionnaire.

[0096] 2. Data preprocessing: Filter, denoise, and standardize the collected multimodal data to form clean data;

[0097] Data preprocessing includes: EEG signal filtering: EEG signal X EEG (t) Use a bandpass filter H(f) with a filtering frequency range of [0.5,50] Hz:

[0098]

[0099] in, and Respectively represent Fourier transform and inverse transform, f represents frequency; Y EEG (t) Filtered EEG signal;

[0100] Speech signal preprocessing: AUDIO (t) Extract Mel frequency cepstrum coefficients:

[0101]

[0102] Among them, C k : kth order cepstrum coefficient; X n : filter energy value; N: number of filters.

[0103] Data standardization: For all modal data X = {x 1 ,x 2 ,…,x m}Perform zero mean standardization:

[0104]

[0105] Among them, x i is the original data, μ is the mean, σ is the standard deviation, Z i is the standardized data.

[0106] 3. Feature extraction: Extract key features from each modality data to form EEG feature vectors F EEG , heart rate feature vector F HR 、Image feature vector F M RI and questionnaire feature vector F Q ;

[0107] Feature extraction includes:

[0108] EEG signal frequency domain features F EEG The steps include: calculating the power spectrum density of the EEG signal and extracting the power of the δ, θ, α, and β bands:

[0109]

[0110] Among them, P δ ,P θ ,P α ,P β are the power values ​​of each band, |X(f)|2 2 is the power spectral density of the signal; X(f) is the spectrum of the EEG signal; f is the frequency;

[0111] EEG signal frequency domain features F EEG =[P δ ,P θ ,P α ,P β ];

[0112] Image feature F MRI The following steps are included: Segment the image into k brain regions R 1 ,R 2 ,…,R k ;

[0113] Extract brain region volumes:

[0114] Where V i : the volume of the ith brain region; ρ(x,y,z): the pixel value of each pixel.

[0115] Image feature F MRI =[V 1 ,V 2 ,…,V i ];

[0116] Extract the heart rate feature vector F HR Includes the following steps: Average heart rate: where μ HR : average heart rate; T: duration of signal acquisition; X HR (t): heart rate signal;

[0117] Heart Rate Variability:

[0118]

[0119] Among them, HRV LF ,HRV HF : Heart rate variability low-frequency and high-frequency power;

[0120] X HR (f): spectrum of heart rate signal;

[0121] Heart rate feature vector F HR =[μ HR ,HRV LF ,HRV HF ];

[0122] Extract the questionnaire feature vector F Q The steps are as follows: Calculate the total score:

[0123]

[0124] Where, S: total score of the questionnaire; W: i : The weight of the i-th item; Q i : The score of the ith item.

[0125] Calculate sub-score: S depression =∑ i∈D W i Q i ,S anxiety =∑ i∈A W i Q i , where S depression ,S anxiety : Depression and anxiety scores; D, A: A collection of questions related to depression and anxiety;

[0126] Questionnaire feature vector F Q =[S,S depression ,S anxiety ];

[0127] 4. Feature fusion: The feature vectors of all modalities are weighted and fused to generate a unified feature vector V:

[0128] Feature fusion formula:

[0129] Among them, w EEG 、w HR 、w MRI 、w Q is the weight coefficient of each mode, satisfying Σw i =1.

[0130] 5. Classification model training and application: Train and classify unified feature vectors based on deep learning models, and output the confidence level of each mental disorder category;

[0131] The classification model uses a deep learning structure and is trained based on the following cross entropy loss function:

[0132]

[0133] Where m is the number of samples, K is the number of mental disorder categories, and y ik is the true category label, is the k-th class probability predicted by the model, θ is the model parameter, and L(θ) is the loss function.

[0134] The classification model is a multi-layer perceptron (MLP), including:

[0135] Input layer: input feature vector Hidden layer: using ReLU activation function:

[0136] h (l) =max(0,W (l) h (l-1) +b (l) )

[0137] Among them, W (l) is the weight matrix, b (l) is bias;

[0138] Output layer: Output category probability through Softmax function:

[0139]

[0140] The model is optimized by introducing a regularization term:

[0141] L′(θ)=L(θ)+λ||θ|| 2

[0142] Among them, λ is the regularization coefficient and ||θ|| is the L2 norm of the parameter.

[0143] 6. Result interpretation: Generate a diagnosis report based on the classification results, indicate high-risk and medium-risk categories, and provide treatment recommendations. Calculate the confidence level based on the probability output by the model:

[0144]

[0145] Among them, C k is the confidence of the kth category, when C k ≥75%, marked as high risk category; when 50%≤C k<75%, marked as medium risk category.

[0146] 7. This method is suitable for hospital outpatient screening, mental health assessment equipment and telemedicine platforms, and is used to quickly identify multiple categories of mental disorders such as depression, anxiety and schizophrenia.

[0147] The following is a specific example of the technical content in conjunction with several embodiments;

[0148] Example 1: Identification of depression based on EEG signals and psychological questionnaires;

[0149] Patient Wang, a 35-year-old male, went to the hospital for a mental status assessment due to long-term depression, loss of interest, and sleep disorders. The doctor arranged for him to undergo an EEG test and a PHQ-9 depression scale test.

[0150] 1. Data collection: EEG signal data: The patient’s EEG signals were collected using an electroencephalograph with a sampling frequency of 256 Hz and a duration of 5 minutes.

[0151] After preprocessing, the filter frequency range is [0.5, 50] Hz. Psychological questionnaire data: PHQ-9 scores are as follows: Q = [3, 2, 3, 3, 2, 1, 2, 3, 3], and the total score is:

[0152] The scores for the core depression questions (questions 1 and 2) are: S core =Q 1 +Q 2 =3+2=5.

[0153] 2. Feature extraction: EEG signal feature extraction: Extract the power of each frequency band through power spectrum analysis: P δ =0.32,P θ =0.28,P α =0.18,P β =0.22;

[0154] The EEG feature vector is: V EEG =[0.32,0.28,0.18,0.22];

[0155] Questionnaire feature extraction: Questionnaire feature vector is: V PHQ =[22,5];

[0156] 3. Feature fusion: Use the weighted fusion formula to generate a unified feature vector: V fusion =w EEG ·V EEG +w PHQ ·V PHQ ;

[0157] Among them, wE EG=0.6,w PHQ =0.4.

[0158] The calculation is as follows: V fusion =0.6·[0.32,0.28,0.18,0.22]+0.4·[22,5]V fusion =[0.192,0.168,0.108,0.132]+[8.8,2]=[8.992,2.168,0.108,0.132].

[0159] 4. Classification and diagnosis: V fusion Input the deep learning model (MLP), and the model outputs the following category probabilities:

[0160] in: normal; Moderate depression; Severe depression.

[0161] Calculate confidence:

[0162] Diagnostic conclusion: Based on the confidence level, the patient is labeled as a medium-risk category and the doctor recommends further evaluation.

[0163] Example 2: Anxiety screening based on heart rate and imaging data;

[0164] Patient Li, a 42-year-old female, went to the hospital due to work pressure, long-term tension, irritability, palpitations and insomnia. The doctor arranged a heart rate variability test and MRI examination.

[0165] 1. Data collection: Heart rate data: Use a heart rate monitor with a sampling frequency of 1 Hz and record for 5 minutes, with a total of 300 heart rate data points.

[0166] The statistical results are as follows:

[0167] Average heart rate:

[0168] Low frequency / high frequency power ratio (LF / HF): HR LF / HF =1.5;

[0169] Image data: MRI is segmented into 116 brain regions, and the brain volume is calculated (assuming that some results are as follows): V MRI =[0.35,0.42,0.37,…,0.25].

[0170] 2. Feature extraction:

[0171] Heart rate features: Extract the average heart rate and LF / HF ratio to form a feature vector: VHR =[85,1.5];

[0172] Image features: The brain volume feature vector is: V MRI =[0.35,0.42,0.37,…,0.25];

[0173] 3. Feature fusion: Use weighted fusion formula: V fusion =w HR ·V HR +w MRI ·V MRI ;

[0174] where w HR =0.5,w MRI = 0.5. The unified eigenvector is calculated.

[0175] 4. Classification and diagnosis: Input fusion feature vector V fusion To the classification model, the model outputs the following category probabilities:

[0176] Calculate the anxiety confidence level:

[0177] Diagnostic conclusion: The patient was labeled as a medium-risk category and intervention with medication and psychotherapy was recommended.

[0178] Example 3: Auxiliary diagnosis of schizophrenia based on speech signals;

[0179] The patient, Mr. Zhang, a 28-year-old male, showed confusion in speech, unclear logic, and hallucinations. The doctor arranged for speech analysis and psychological scale tests.

[0180] 1. Data collection: voice signal:

[0181] The patient's retelling of a certain passage was collected at a sampling frequency of 16kHz and a duration of 1 minute, with a total of 16,000×60=960,000 data points.

[0182] Extract Mel Frequency Cepstral Coefficients (MFCC): V MFCC =[12.3,10.8,9.6,…,8.7];

[0183] Psychological scale: PANSS scale (schizophrenia assessment) score: Q = [5, 4, 5, 6, 7, 6, 5];

[0184] Total score:

[0185] 2. Feature extraction:

[0186] Speech features: Extract the first 13 MFCC features: VMFCC =[12.3,10.8,9.6,…,8.7];

[0187] Questionnaire features: The questionnaire feature vector is: V PANSS =

[38] ;

[0188] 3. Feature Fusion:

[0189] V fusion =0.6 V MFCC +0.4 V PANSS ;

[0190] 4. Classification and diagnosis:

[0191] Input the feature vector to the model and output the following category probabilities:

[0192]

[0193] Calculate confidence:

[0194] Diagnostic conclusion: The patient was labeled as a high-risk category and immediate drug intervention and hospitalization were recommended.

[0195] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 invention.

Claims

1. A method for identifying multiple categories of mental disorders, characterized in that The following steps are involved: Multimodal data collection: Collect multimodal data of patients, including physiological data, including EEG signals X EEG (t), heart rate X HR (t) and skin conductance X EDA (t), behavioral data, behavioral data includes facial expression X FACE and speech signal X AUDIO (t), medical imaging data X MRI , and psychological questionnaire data X Q = {Q1,Q2,…,Q n }; Data preprocessing: filtering, denoising, and standardizing the collected multimodal data to form clean data; Feature extraction: Extract key features from each modality data to form EEG feature vectors F EEG , heart rate feature vector F HR 、Image feature vector F M RI and questionnaire feature vector F Q ; feature Fusion: The feature vectors of all modalities are weightedly fused to generate a unified feature vector V; Classification model training and application: Train and classify unified feature vectors based on deep learning models, and output the confidence level of each mental disorder category; Result interpretation: A diagnostic report is generated based on the classification results, indicating high-risk and medium-risk categories and providing treatment recommendations.

2. The method according to claim 1, characterized in that The data collection specifically includes: Physiological data collection: EEG signals X EEG (t): sampling frequency f s =500Hz, where t represents time and the signal unit is microvolt (μV); Heart RateX HR (t): sampling frequency f s =1Hz, unit is beats per minute (BPM); Skin ConductanceX EDA (t): The unit is Siemens (μS), the sampling resolution is 10 -3 μS; Behavioral data collection: Facial expression sequence X FACE : Obtained through video, frame rate f video =30FPS; voice signal X AUDIO (t): sampling frequency f s =16kHz, unit is amplitude value; Imaging data acquisition: Magnetic resonance imaging dataX MRI , resolution 1×1×1mm 3 ; Psychological questionnaire data: questionnaire score vector X Q ={q1,q2,…,q n }, where q i ∈[0,4] represents the score of the i-th item, and n is the total number of questions in the questionnaire.

3. A method for identifying multiple categories of mental disorders according to claim 1, characterized in that: The data preprocessing includes: EEG signal filtering: EEG (t) Use a bandpass filter H(f) with a filtering frequency range of [0.5,50] Hz: in, and Respectively represent Fourier transform and inverse transform, f represents frequency; Y EEG (t) Filtered EEG signal; Speech signal preprocessing: AUDIO (t) Extract Mel frequency cepstrum coefficients: Among them, C k : kth order cepstrum coefficient; X n : filter energy value; N: number of filters. Data standardization: For all modal data X = {x1, x2, ..., x m }Perform zero mean standardization: Among them, x i is the original data, μ is the mean, σ is the standard deviation, Z i is the standardized data.

4. A method for identifying multiple categories of mental disorders according to claim 1, characterized in that: The feature extraction comprises: EEG signal frequency domain features F EEG The steps include: calculating the power spectrum density of the EEG signal and extracting the power of the δ, θ, α, and β bands: Among them, P δ ,P θ ,P α ,P β are the power values ​​of each band, |X(f)| 2 is the power spectral density of the signal; X(f) is the spectrum of the EEG signal; f is the frequency; EEG signal frequency domain features F EEG =[P δ ,P θ ,P α ,P β ]; Image feature F MRI The following steps are included: Segment the image into k brain regions R1, R2, …, R k ; Extract brain region volumes: Where V i : the volume of the ith brain region; ρ(x,y,z): the pixel value of each pixel. Image feature F MRI =[V1,V2,…,V i ]; Extract the heart rate feature vector F HR Includes the following steps: Average heart rate: where μ HR : average heart rate; T: duration of signal acquisition; X HR (t): heart rate signal; Heart Rate Variability: Among them, HRV LF ,HRV HF : Heart rate variability low-frequency and high-frequency power; X HR (f): spectrum of heart rate signal; Heart rate feature vector F HR =[μ HR ,HRV LF ,HRV HF ]; Extract the questionnaire feature vector F Q The steps are as follows: Calculate the total score: Where, S: total score of the questionnaire; W: i : The weight of the i-th item; Q i : The score of the ith item. Calculate sub-score: S depression =Σ i∈D W i Q i ,S anxiety =∑ i∈A W i Q i , where S depression ,S anxiety : Depression and anxiety scores; D, A: A collection of questions related to depression and anxiety; Questionnaire feature vector F Q =[S,S depression ,S anxiety ].

5. A method for identifying multiple categories of mental disorders according to claim 1, characterized in that: The feature vectors of all modes are weighted fused to generate a unified feature vector V: Feature fusion formula: Among them, w EEG 、w HR 、w MRI 、w Q is the weight coefficient of each mode, satisfying ∑w i =1.

6. A method for identifying multiple categories of mental disorders according to claim 1, characterized in that: The classification model uses a deep learning structure and is trained based on the following cross entropy loss function: Where m is the number of samples, K is the number of mental disorder categories, and y ik is the true category label, is the k-th class probability predicted by the model, θ is the model parameter, and L(θ) is the loss function.

7. A method for identifying multiple categories of mental disorders according to claim 6, characterized in that: The model is optimized by introducing a regularization term: L′(θ)=L(θ)+λ||θ|| 2 , Among them, λ is the regularization coefficient and ||θ|| is the L2 norm of the parameter.

8. A method for identifying multiple categories of mental disorders according to claim 6, characterized in that: The classification model is a multi-layer perceptron, including: Input layer: input feature vector Hidden layer: using ReLU activation function: h (l) =max(0,W (l) h (l-1) +b (l) ), Among them, W (l) is the weight matrix, b (l) is bias; Output layer: Output category probability through Softmax function:

9. A method for identifying multiple categories of mental disorders according to claim 1, characterized in that: Calculate the confidence based on the probability output by the model: Among them, C k is the confidence of the kth category, when C k ≥75%, marked as high risk category; when 50%≤C k <75%, marked as medium risk category.

10. A method for identifying multiple categories of mental disorders according to claim 1, characterized in that: The method is applicable to hospital outpatient screening, mental health assessment equipment and telemedicine platforms, and is used to quickly identify multiple categories of mental disorders such as depression, anxiety and schizophrenia.

Images