CN117747064A - Depression clinical decision method and system based on AI - Google Patents
Depression clinical decision method and system based on AI Download PDFInfo
- Publication number
- CN117747064A CN117747064A CN202311664055.8A CN202311664055A CN117747064A CN 117747064 A CN117747064 A CN 117747064A CN 202311664055 A CN202311664055 A CN 202311664055A CN 117747064 A CN117747064 A CN 117747064A
- Authority
- CN
- China
- Prior art keywords
- data
- module
- strategy
- feature
- depression
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000011282 treatment Methods 0.000 claims abstract description 93
- 238000004458 analytical method Methods 0.000 claims abstract description 79
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 73
- 230000004927 fusion Effects 0.000 claims abstract description 62
- 230000008921 facial expression Effects 0.000 claims abstract description 60
- 238000011269 treatment regimen Methods 0.000 claims abstract description 58
- 230000008451 emotion Effects 0.000 claims abstract description 56
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 35
- 208000020401 Depressive disease Diseases 0.000 claims abstract description 28
- 238000004088 simulation Methods 0.000 claims abstract description 28
- 238000003066 decision tree Methods 0.000 claims abstract description 22
- 238000007637 random forest analysis Methods 0.000 claims abstract description 19
- 238000007781 pre-processing Methods 0.000 claims abstract description 18
- 238000005516 engineering process Methods 0.000 claims abstract description 17
- 230000007787 long-term memory Effects 0.000 claims abstract description 13
- 230000008859 change Effects 0.000 claims abstract description 10
- 230000008909 emotion recognition Effects 0.000 claims abstract description 10
- 238000007405 data analysis Methods 0.000 claims abstract description 3
- 230000000694 effects Effects 0.000 claims description 44
- 238000003909 pattern recognition Methods 0.000 claims description 43
- 230000006399 behavior Effects 0.000 claims description 35
- 238000011156 evaluation Methods 0.000 claims description 34
- 238000005457 optimization Methods 0.000 claims description 34
- 230000004044 response Effects 0.000 claims description 31
- 238000000605 extraction Methods 0.000 claims description 29
- 238000012545 processing Methods 0.000 claims description 27
- 238000012360 testing method Methods 0.000 claims description 16
- 238000000342 Monte Carlo simulation Methods 0.000 claims description 13
- 230000001815 facial effect Effects 0.000 claims description 13
- 230000002996 emotional effect Effects 0.000 claims description 12
- 230000014509 gene expression Effects 0.000 claims description 12
- 238000012544 monitoring process Methods 0.000 claims description 12
- 238000012795 verification Methods 0.000 claims description 12
- 238000013135 deep learning Methods 0.000 claims description 11
- 230000010354 integration Effects 0.000 claims description 11
- 230000002787 reinforcement Effects 0.000 claims description 11
- 238000012706 support-vector machine Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 10
- 238000012300 Sequence Analysis Methods 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 239000000284 extract Substances 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000015654 memory Effects 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 8
- 238000011425 standardization method Methods 0.000 claims description 8
- 125000004122 cyclic group Chemical group 0.000 claims description 7
- 238000011217 control strategy Methods 0.000 claims description 6
- 238000013210 evaluation model Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000012952 Resampling Methods 0.000 claims description 5
- 230000008901 benefit Effects 0.000 claims description 5
- 210000001097 facial muscle Anatomy 0.000 claims description 3
- 230000036541 health Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 claims description 2
- 230000036651 mood Effects 0.000 claims 1
- 238000013473 artificial intelligence Methods 0.000 abstract description 5
- 230000001225 therapeutic effect Effects 0.000 description 7
- 238000003745 diagnosis Methods 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000006403 short-term memory Effects 0.000 description 3
- 230000001360 synchronised effect Effects 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000013145 classification model Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000012731 temporal analysis Methods 0.000 description 2
- 238000000700 time series analysis Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 206010061818 Disease progression Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 230000005750 disease progression Effects 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000008713 feedback mechanism Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000004630 mental health Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000009323 psychological health Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000004797 therapeutic response Effects 0.000 description 1
Landscapes
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention relates to the technical field of artificial intelligence, in particular to an AI-based depression clinical decision method and system, comprising the following steps: based on the depression patients, the data preprocessing technology is adopted to clean and synchronize voice, video, text and physiological signals, so as to generate multi-mode patient data. According to the invention, personalized depression treatment prediction and assessment are realized through multi-mode data analysis and feature fusion, the emotion recognition accuracy of voice and video data is improved by combining a long-term memory network and a three-dimensional convolutional neural network, the robustness of a prediction model is enhanced by a random forest and gradient lifting decision tree, the treatment strategy is optimized by a depth strategy gradient algorithm, the success rate is continuously improved by matching with a simulation platform, accurate interpretation of a patient is realized by real-time facial expression tracking and emotion analysis, the dynamic adjustment of clinical decisions is realized by an online learning algorithm, the state change of the patient is responded timely, and the individuation and the accuracy of treatment are improved.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to an AI-based depression clinical decision method and system.
Background
Artificial intelligence is a field of research on how computer systems exhibit intelligent behavior. It encompasses a number of sub-fields of machine learning, deep learning, natural language processing, computer vision, and the like. AI technology is widely used in the medical field, particularly in clinical decision support systems. AI is capable of processing and analyzing large amounts of medical data, identifying patterns, assisting doctors in diagnosing, predicting disease progression, providing personalized treatment regimens, and improving patient care.
The AI-based depression clinical decision method is a method for diagnosis, prediction and treatment support of depression by combining medical data and psychological knowledge by utilizing an artificial intelligence technology. The method aims at assisting doctors in diagnosing depression by analyzing multidimensional data such as physiology, psychology and behavior of patients, predicting the disease development trend of the patients, making personalized treatment plans, providing timely intervention and support, improving the life quality of the patients and reducing the psychological health risks of the patients suffering from depression. The AI-based depression clinical decision method provides more accurate, efficient and personalized depression diagnosis and treatment schemes through the support of artificial intelligence technology, and provides better medical service and mental health support for patients.
The existing depression treatment decision-making method mostly depends on subjective experience of doctors and linear statistical models, and lacks deep and dynamic analysis on emotion and behavior patterns of patients. The methods have low efficiency when processing complex multi-mode data, and often cannot realize deep fusion of data features, so that the recognition accuracy and the prediction accuracy are limited. In predicting the response of the treatment, the traditional method fails to effectively utilize the machine learning algorithm to mine the potential nonlinear relationship of the data, so that the treatment scheme often lacks individualization and dynamic adaptation capability. Furthermore, the lack of integration of an immediate feedback mechanism makes it difficult to adjust the treatment regimen in real time, which affects the maximization of the therapeutic effect to some extent. Therefore, the existing method has limitation in the aspect of multidimensional complexity of treating depression, and the potential of clinical big data and advanced algorithms in treatment individuation and treatment effect improvement cannot be fully exerted.
Disclosure of Invention
The invention aims to solve the defects in the prior art, and provides an AI-based depression clinical decision method and an AI-based depression clinical decision system.
In order to achieve the above purpose, the present invention adopts the following technical scheme: an AI-based depression clinical decision method comprising the steps of:
S1: based on the depression patient, adopting a data preprocessing technology to clean and synchronize voice, video, text and physiological signals to generate multi-mode patient data;
s2: based on the multi-modal patient data, adopting a long-short-term memory network and a three-dimensional convolutional neural network to analyze voice and video data, and carrying out feature fusion to generate a psychological behavior pattern recognition result;
s3: based on the psychological behavior pattern recognition result, adopting a random forest and gradient lifting decision tree to predict the treatment response of the patient, evaluating the treatment effect, and generating a depression prediction result and a treatment effect evaluation report;
s4: based on the treatment effect evaluation report, adopting a depth deterministic strategy gradient algorithm to optimize a strategy, and testing on a simulation platform to generate an optimized depression treatment strategy;
s5: based on video data in the multi-mode patient data, adopting a convolutional neural network and cyclic neural network mixed model to track facial expressions and analyze emotional states, and generating facial expression recognition and emotion analysis reports;
s6: based on the optimized depression treatment strategy, facial expression recognition and emotion analysis report, adopting an online learning algorithm to adjust clinical decisions in real time, and generating optimized clinical decisions and dynamic adjustment strategies;
The multi-modal patient data is specifically a patient information set subjected to noise reduction, outlier processing and time stamp alignment, the psychological behavior pattern recognition result comprises an emotion state label, language expression habit and limb language characteristics, the depression prediction result is specifically probability distribution of future health states of the patient, the treatment effect evaluation report comprises benefit comparison of multiple treatment schemes, and the facial expression recognition and emotion analysis report comprises facial muscle activity data and emotion change-based time sequence analysis.
As a further scheme of the invention, based on the depression patient, the data preprocessing technology is adopted to clean and synchronize voice, video, text and physiological signals, and the steps for generating the multi-mode patient data are specifically as follows:
s101: performing frequency domain analysis by adopting Fourier transform based on original multi-mode data of a depression patient, and processing missing values by mean filling and nearest neighbor interpolation to generate preprocessed multi-mode data;
s102: resampling and time alignment are carried out by adopting a time sequence analysis method based on the preprocessed multi-mode data, so as to generate time-synchronous multi-mode data;
S103: based on the time-synchronous multi-mode data, extracting key features by adopting a Mel frequency cepstrum coefficient and a deep learning feature automatic encoder to generate a multi-mode feature data set;
s104: and based on the multi-modal feature data set, adopting a weighted fusion strategy to fuse multi-modal data and generating fused multi-modal feature data.
As a further scheme of the invention, based on the multi-modal patient data, a long-term memory network and a three-dimensional convolutional neural network are adopted to analyze voice and video data, and feature fusion is carried out, so that the step of generating a psychological behavior pattern recognition result is specifically as follows:
s201: based on the fused multi-mode characteristic data, extracting dynamic time characteristics of voice by adopting a long-short-period memory network, and generating voice time sequence characteristics;
s202: based on the fused multi-mode characteristic data, extracting the spatial characteristics and dynamic changes of the video by adopting a three-dimensional convolutional neural network, and generating the space-time characteristics of the video;
s203: based on the voice time sequence features and the video space-time features, performing depth feature fusion by using feature cascading and fusion strategies to generate voice video depth feature fusion data;
S204: and carrying out pattern recognition by adopting a support vector machine based on the voice video depth feature fusion data to generate a psychological behavior pattern recognition result.
As a further scheme of the present invention, based on the psychological behavior pattern recognition result, a random forest and gradient lifting decision tree is adopted to predict the therapeutic response of the patient, and evaluate the therapeutic effect, and the steps of generating the predicted result of depression and the evaluation report of therapeutic effect are specifically as follows:
s301: based on the psychological behavior pattern recognition result, ranking the feature importance by adopting a random forest algorithm, analyzing the change before and after treatment, and generating baseline feature analysis data;
s302: constructing a prediction model by adopting a gradient lifting decision tree based on the baseline characteristic analysis data, and performing treatment response prediction to generate treatment response prediction data;
s303: based on the treatment response prediction data, performing model verification and tuning, and using k-fold cross verification to ensure generalization capability of the model to generate a treatment effect evaluation model;
s304: and analyzing the model performance by adopting a confusion matrix and a receiver operation characteristic curve based on the treatment effect evaluation model, and generating a depression predicted result and a treatment effect evaluation report.
As a further scheme of the invention, based on the treatment effect evaluation report, a depth deterministic strategy gradient algorithm optimization strategy is adopted, and a simulation platform is used for testing, and the steps for generating the optimized depression treatment strategy are specifically as follows:
s401: based on the treatment effect evaluation report, performing scale processing on the data by adopting a Z-score standardization method, and generating normalized evaluation data;
s402: based on the normalized evaluation data, training a strategy network by adopting a depth deterministic strategy gradient algorithm to generate a primarily optimized treatment strategy model;
s403: based on the primarily optimized treatment strategy model, performing Monte Carlo simulation in a simulation environment, collecting decision performance, and generating simulation test data;
s404: and based on the simulation test data, the DDPG algorithm refinement strategy is applied again, and the optimized depression treatment strategy is generated.
As a further scheme of the present invention, based on the video data in the multi-modal patient data, a convolutional neural network and cyclic neural network hybrid model is adopted to track facial expressions and analyze emotional states, and the steps of generating facial expression recognition and emotion analysis reports specifically include:
S501: based on video data in the multi-mode patient data, performing frame extraction and gray scale processing through an OpenCV tool to generate preprocessed video frame data;
s502: based on the preprocessed video frame data, extracting spatial features by adopting a convolutional neural network to generate facial feature data;
s503: based on the facial feature data, analyzing the expression sequence by applying a long-term memory network to generate facial expression dynamic data;
s504: and based on the facial expression dynamic data, integrating facial feature data and facial expression dynamic data, and generating facial expression recognition and emotion analysis reports by adopting an emotion recognition algorithm.
As a further scheme of the invention, based on the optimized depression treatment strategy and facial expression recognition and emotion analysis report, the clinical decision is adjusted in real time by adopting an online learning algorithm, and the steps of generating the optimized clinical decision and dynamic adjustment strategy are specifically as follows:
s601: based on the optimized depression treatment strategy, parameters are adjusted by using an online learning method, and initially adjusted clinical decision parameters are generated;
s602: optimizing decision parameters based on the facial expression recognition and emotion analysis report in combination with reinforcement learning technology to generate emotion-adjusted clinical decision parameters;
S603: based on the emotion-adjusted clinical decision parameters, real-time monitoring and dynamic adjustment are carried out, and self-adaptive control strategies are used for generating real-time monitoring adjustment data;
s604: based on the real-time monitoring and adjusting data, final online learning optimization is executed, real-time performance and individuation of decisions are ensured, and optimized clinical decisions and dynamic adjusting strategies are generated.
An AI-based depression clinical decision system is used for executing the AI-based depression clinical decision method, and comprises a data preprocessing module, a multi-mode feature extraction module, a feature fusion and mode identification module, a treatment strategy optimization module, an emotion analysis module, a decision parameter optimization module and a system integration and optimization module.
As a further scheme of the invention, the data preprocessing module adopts a Fourier transform algorithm to perform frequency domain analysis based on original multi-mode data of a patient suffering from depression, and utilizes a mean value filling method and a nearest neighbor interpolation method to process a missing value so as to generate preprocessed multi-mode data;
the multi-modal feature extraction module performs complex sampling and time alignment by adopting a time sequence analysis method based on the preprocessed multi-modal data, automatically extracts key features by utilizing a Mel frequency cepstrum coefficient and a deep learning feature automatic encoder, and generates a time-synchronous multi-modal feature data set;
The feature fusion and pattern recognition module utilizes a time-synchronous multi-modal feature data set, utilizes a weighted fusion strategy to fuse data, analyzes patterns through a support vector machine algorithm, and generates fused multi-modal feature data and a psychological behavior pattern recognition result;
the treatment strategy optimization module sorts the importance of the features by adopting a random forest algorithm based on the psychological behavior pattern recognition result, and establishes a prediction model by utilizing a gradient lifting decision tree algorithm to generate baseline feature analysis data and treatment response prediction data;
the emotion analysis module extracts voice and video features by using a long-term memory network algorithm and a three-dimensional convolutional neural network according to the fused multi-modal feature data, analyzes the voice and video features by applying an emotion recognition algorithm, and generates voice video depth feature fusion data and facial expression recognition and emotion analysis reports;
the decision parameter tuning module is used for adjusting parameters based on treatment response prediction data and facial expression recognition and emotion analysis reports by utilizing an online learning method and a reinforcement learning technology to generate preliminarily adjusted clinical decision parameters and emotion adjusted clinical decision parameters;
The system integration and optimization module analyzes model performance by adopting a confusion matrix and a receiver operation characteristic curve, performs large-scale processing on data by a Z score standardization method, trains a strategy network by utilizing a depth deterministic strategy gradient algorithm, performs Monte Carlo simulation in a simulation environment, collects decision expression, applies a depth deterministic strategy gradient algorithm refinement strategy again, and generates an optimized treatment strategy and an optimized depression treatment strategy.
As a further scheme of the invention, the data preprocessing module comprises a frequency domain analysis sub-module, a missing value processing sub-module and a data normalization sub-module;
the multi-mode feature extraction module comprises a time alignment sub-module, a key feature extraction sub-module and a feature data set construction sub-module;
the feature fusion and pattern recognition module comprises a data fusion sub-module, a pattern recognition sub-module and a feature cascade sub-module;
the treatment strategy optimization module comprises a feature importance ordering sub-module, a prediction model construction sub-module and a model verification and tuning sub-module;
the emotion analysis module comprises a voice feature extraction sub-module, a video feature extraction sub-module and a feature fusion sub-module;
The decision parameter tuning module comprises a feature selection sub-module, a model selection sub-module, a super parameter tuning sub-module and a model evaluation sub-module;
the system integration and optimization module comprises a performance analysis sub-module, a data scale sub-module, a strategy network training sub-module, a Monte Carlo simulation sub-module and a strategy refinement sub-module.
Compared with the prior art, the invention has the advantages and positive effects that:
according to the invention, through multi-mode data comprehensive analysis and multi-level feature fusion, a more accurate and personalized depression treatment prediction and assessment system is provided. The accuracy of voice and video data in the psychological behavior pattern recognition is improved by using the long-short-term memory network and the three-dimensional convolutional neural network, so that emotion recognition and analysis are more detailed and comprehensive. The combination of random forests and gradient boosting decision trees is used to predict treatment response, enhancing the robustness and interpretation of the predictive model. The depth deterministic strategy gradient algorithm optimizes the treatment strategy, provides a simulation platform for testing and verification, and can continuously improve the treatment success rate. Real-time tracking of facial expressions and emotional state analysis enable more accurate interpretation of patient feedback. The method realizes the dynamic adjustment of clinical decisions through an online learning algorithm, ensures that the treatment scheme can timely respond to the tiny change of the state of a patient, and improves the individuation and the accuracy of treatment.
Drawings
FIG. 1 is a schematic workflow diagram of the present invention;
FIG. 2 is a S1 refinement flowchart of the present invention;
FIG. 3 is a S2 refinement flowchart of the present invention;
FIG. 4 is a S3 refinement flowchart of the present invention;
FIG. 5 is a S4 refinement flowchart of the present invention;
FIG. 6 is a S5 refinement flowchart of the present invention;
FIG. 7 is a S6 refinement flowchart of the present invention;
FIG. 8 is a system flow diagram of the present invention;
FIG. 9 is a schematic diagram of a system framework of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
Example 1
Referring to fig. 1, the present invention provides a technical solution: an AI-based depression clinical decision method comprising the steps of:
s1: based on the depression patient, adopting a data preprocessing technology to clean and synchronize voice, video, text and physiological signals to generate multi-mode patient data;
s2: based on multi-mode patient data, adopting a long-term memory network and a three-dimensional convolutional neural network to analyze voice and video data, and carrying out feature fusion to generate a psychological behavior mode recognition result;
s3: based on the psychological behavior pattern recognition result, adopting a random forest and gradient lifting decision tree to predict the treatment response of the patient, evaluating the treatment effect, and generating a depression prediction result and a treatment effect evaluation report;
s4: based on the treatment effect evaluation report, adopting a depth deterministic strategy gradient algorithm to optimize the strategy, and testing on a simulation platform to generate an optimized depression treatment strategy;
s5: based on video data in the multi-mode patient data, tracking facial expressions and analyzing the emotional states by adopting a convolutional neural network and cyclic neural network mixed model to generate facial expression recognition and emotion analysis reports;
S6: based on the optimized depression treatment strategy, facial expression recognition and emotion analysis report, adopting an online learning algorithm to adjust clinical decisions in real time, and generating an optimized clinical decision and a dynamic adjustment strategy;
the multi-modal patient data is specifically a patient information set subjected to noise reduction, outlier processing and time stamp alignment, the psychological behavior pattern recognition result comprises an emotion state label, language expression habit and limb language characteristics, the depression prediction result is specifically probability distribution of future health states of the patient, the treatment effect evaluation report comprises benefit comparison of a multi-treatment scheme, and the facial expression recognition and emotion analysis report comprises facial muscle activity data and emotion change-based time series analysis.
Firstly, the voice, video, text and physiological signals are cleaned and synchronized through a data preprocessing technology to generate multi-mode patient data, so that the quality and usability of the data are greatly improved, and a solid foundation is provided for subsequent analysis and prediction. And secondly, the method adopts a long-term memory network and a three-dimensional convolutional neural network to analyze voice and video data, performs feature fusion, and generates a psychological behavior pattern recognition result.
In addition, the method also adopts a random forest and gradient lifting decision tree to predict the treatment response of the patient, evaluates the treatment effect and generates a depression prediction result and a treatment effect evaluation report. Meanwhile, the method also adopts a depth deterministic strategy gradient algorithm optimization strategy, and tests are carried out on a simulation platform to generate an optimized depression treatment strategy.
Finally, the method also adopts a convolutional neural network and cyclic neural network mixed model to track facial expressions and analyze emotional states, and generates facial expression recognition and emotion analysis reports. In general, the clinical decision method for the depression based on the AI can improve the diagnosis accuracy, optimize the treatment strategy, improve the treatment effect, provide more comprehensive patient information for doctors, and have important significance for improving the treatment effect of depression patients.
Referring to fig. 2, based on the depression patient, the steps of cleaning and synchronizing the voice, video, text and physiological signals by adopting the data preprocessing technology to generate the multi-modal patient data are specifically as follows:
s101: performing frequency domain analysis by adopting Fourier transform based on original multi-mode data of a depression patient, and processing missing values by mean filling and nearest neighbor interpolation to generate preprocessed multi-mode data;
s102: resampling and time alignment are carried out by adopting a time sequence analysis method based on the preprocessed multi-modal data, so as to generate time-synchronous multi-modal data;
s103: based on time-synchronous multi-modal data, extracting key features by adopting a Mel frequency cepstrum coefficient and a deep learning feature automatic encoder to generate a multi-modal feature data set;
s104: based on the multi-modal feature data set, adopting a weighted fusion strategy to fuse multi-modal data and generating fused multi-modal feature data.
The original multi-modal data is frequency domain analyzed using fourier transforms. By converting the time domain signal into a frequency domain signal, the frequency characteristics of the data can be better understood. Next, the missing values are processed using mean padding and nearest neighbor interpolation. For missing data points, the missing values can be filled in by calculating the average value of the adjacent data points so as to ensure the integrity of the data.
Resampling and time alignment of the preprocessed multimodal data using time series analysis. Since data of different modalities may have different sampling rates and time scales, it is necessary to unify the data on a common time axis. By resampling and time alignment, it can be ensured that the data of the various modalities are synchronized in time.
Key features are extracted using mel-frequency cepstral coefficients (MFCCs) and a deep learning feature auto-encoder. MFCC is a feature extraction method commonly used for speech signals, and captures the spectral characteristics of the speech signals. While deep learning feature auto-encoders learn higher level abstract features. By combining the MFCC and the deep learning feature auto-encoder, more representative key features can be extracted from the multi-modal data.
And fusing the multi-modal feature data set by using a weighted fusion strategy. Because the data of different modes has different information amounts and importance, a weight can be allocated to each mode to reflect the contribution degree of the data in the final fusion result. Through a weighted fusion strategy, the characteristics of all modes can be effectively integrated together to generate the fused multi-mode characteristic data.
Referring to fig. 3, based on multi-modal patient data, the steps of performing voice and video data analysis by using a long-short-term memory network and a three-dimensional convolutional neural network, and performing feature fusion to generate a psycho-behavioral pattern recognition result are specifically as follows:
s201: based on the fused multi-mode characteristic data, extracting the dynamic time characteristic of the voice by adopting a long-period memory network to generate a voice time sequence characteristic;
s202: based on the fused multi-mode characteristic data, extracting the spatial characteristics and dynamic changes of the video by adopting a three-dimensional convolutional neural network, and generating the space-time characteristics of the video;
s203: based on the voice time sequence features and the video space-time features, performing depth feature fusion by using feature cascading and fusion strategies to generate voice video depth feature fusion data;
s204: and carrying out pattern recognition by adopting a support vector machine based on the depth feature fusion data of the voice and video to generate a psychological behavior pattern recognition result.
And extracting the voice time sequence characteristics of the fused multi-mode characteristic data by using a long-short-term memory network (LSTM). LSTM is a recurrent neural network structure capable of capturing long-term dependencies, which can efficiently process voice data with a time dimension. By inputting the fused features into the LSTM network, the dynamic time characteristics of the speech data can be learned and represented as continuous time series feature vectors.
And extracting video space-time characteristics of the fused multi-mode characteristic data by using a three-dimensional convolutional neural network (3D-CNN). The 3D-CNN may take into account both the spatial and temporal dimensions of the video data, thereby better capturing motion information and spatial variations in the video. By inputting the fused features into the 3D-CNN network, the spatio-temporal features of the video data can be learned and represented as continuous video spatio-temporal feature vectors.
And carrying out depth feature fusion on the voice time sequence features and the video space-time features by using feature cascading and fusion strategies. The feature cascade connection refers to connecting features of different modes according to a certain sequence to form a new feature vector. The feature fusion refers to performing operations such as weighted average or maximum pooling on features of different modes to obtain a comprehensive feature representation. Through feature cascading and fusion strategies, features of voice and video can be effectively integrated together to generate voice video depth feature fusion data.
And performing pattern recognition on the audio and video depth feature fusion data by using a Support Vector Machine (SVM) to generate a psychological behavior pattern recognition result. SVM is a commonly used classification algorithm that can separate different classes of data by finding an optimal hyperplane. By training an SVM classifier, the voice video depth feature fusion data can be mapped to a specific psychological behavior pattern, so that the psychological behavior pattern of a depression patient can be identified.
Referring to fig. 4, based on the psychological behavior pattern recognition result, the treatment response prediction of the patient is performed by adopting a random forest and gradient lifting decision tree, and the treatment effect is evaluated, and the steps for generating the depression prediction result and the treatment effect evaluation report are specifically as follows:
s301: based on the psychological behavior pattern recognition result, ranking the feature importance by adopting a random forest algorithm, analyzing the change before and after treatment, and generating baseline feature analysis data;
s302: constructing a prediction model by adopting a gradient lifting decision tree based on the baseline characteristic analysis data, and predicting treatment response to generate treatment response prediction data;
s303: based on the treatment response prediction data, performing model verification and optimization, and using k-fold cross verification to ensure the generalization capability of the model to generate a treatment effect evaluation model;
s304: and analyzing the model performance by adopting the confusion matrix and the receiver operation characteristic curve based on the treatment effect evaluation model, and generating a depression predicted result and a treatment effect evaluation report.
The feature importance was ranked using a random forest algorithm and analyzed for changes before and after treatment. Random forests are an ensemble learning algorithm that can classify or regression predictions by voting or averaging across multiple decision trees. By calculating the relative importance of each feature in the random forest, it can be determined which features have a greater impact on the treatment response. From these baseline signature analysis data, the relationship between different signatures and treatment responses can be understood.
And constructing a prediction model by using a gradient lifting decision tree to predict treatment response. The gradient lifting decision tree is an iterative optimized decision tree algorithm, which can improve the performance of the model by continuously adding new decision trees. By inputting the baseline characteristic analysis data into the gradient-lifting decision tree model, the response of the patient to different treatment methods can be predicted, and treatment response prediction data can be generated.
Model verification and optimization are carried out on the treatment response prediction data. To ensure generalization ability of the model, the performance of the model on different subsets can be evaluated using a k-fold cross-validation method. The accuracy and stability of the model can be further improved by adjusting the hyper-parameters of the model and selecting the appropriate feature subset. Finally, a validated and tuned treatment effect assessment model is generated.
Model performance is analyzed using the confusion matrix and receiver operating characteristic curve (ROC curve) and depression prediction results and treatment effect assessment reports are generated. The confusion matrix may be used to evaluate the accuracy, recall, and precision of the classification model. The ROC curve can be used to evaluate the balance between true and false positive rates of the classification model at different thresholds. By comprehensively analyzing the indexes, a predicted depression result and a treatment effect evaluation report can be obtained, and a reference basis is provided for clinical decision.
Referring to fig. 5, based on the treatment effect evaluation report, a depth deterministic strategy gradient algorithm is adopted to optimize the strategy, and a simulation platform is used for testing, so that the optimized treatment strategy for the depression is specifically generated by the following steps:
s401: based on the treatment effect evaluation report, adopting a Z-score standardization method to carry out scale processing on the data and generating normalized evaluation data;
s402: based on the normalized evaluation data, training a strategy network by adopting a depth deterministic strategy gradient algorithm to generate a primarily optimized treatment strategy model;
s403: based on the primarily optimized treatment strategy model, performing Monte Carlo simulation in a simulation environment, collecting decision performance, and generating simulation test data;
s404: based on the simulation test data, the DDPG algorithm refinement strategy is applied again, and the optimized depression treatment strategy is generated.
In order to eliminate dimensional differences among different features, a Z-score standardization method can be adopted to conduct large-scale processing on data. And converting the value of each characteristic into a Z score corresponding to the value by calculating the mean value and the standard deviation of each sample, thereby realizing the normalization processing of the data.
Next, a depth deterministic strategy gradient algorithm (DDPG) is used to train the strategy network, generating a primarily optimized treatment strategy model. DDPG is an algorithm based on reinforcement learning, which can be used to solve the problem of continuous motion space. In the training process, the normalized evaluation data is used as input, and parameters of the strategy network are updated continuously and iteratively, so that the strategy network can select the optimal action according to the current environment state.
Then, monte Carlo simulation is carried out in a simulation environment, decision expressions are collected, and simulation test data are generated. Monte Carlo simulation is a commonly used random sampling method that can be used to estimate the performance of a system or process. The primarily optimized treatment strategy model can be applied to a simulation environment, and decision-making performance data are collected through multiple times of running simulation experiments. These data can be used to evaluate the efficacy and stability of a treatment strategy.
And finally, based on the simulation test data, the DDPG algorithm refinement strategy is applied again, and the optimized depression treatment strategy is generated. By analyzing the simulation test data, it can be found that the policy network has shortcomings in certain situations. Therefore, the hyper-parameters of the DDPG algorithm can be further adjusted, or the structure of the strategy network is improved, so that the performance and the robustness of the treatment strategy are further improved.
Referring to fig. 6, based on video data in multi-modal patient data, a convolutional neural network and cyclic neural network hybrid model is adopted to track facial expressions and analyze emotional states, and the steps of generating facial expression recognition and emotion analysis reports are specifically as follows:
s501: based on video data in the multi-mode patient data, performing frame extraction and gray scale processing through an OpenCV tool to generate preprocessed video frame data;
S502: based on the preprocessed video frame data, extracting spatial features by adopting a convolutional neural network to generate facial feature data;
s503: based on facial feature data, applying a long-term and short-term memory network to analyze the expression sequence and generating facial expression dynamic data;
s504: based on the facial expression dynamic data, integrating facial feature data and facial expression dynamic data, and generating facial expression recognition and emotion analysis reports by adopting an emotion recognition algorithm.
Frame extraction and gray scale processing are performed on video data in multi-modal patient data using an OpenCV tool. By decomposing the video into successive image frames, each frame may be pre-processed, including graying, scaling, etc., to facilitate subsequent feature extraction and analysis.
Convolutional Neural Networks (CNNs) are employed to extract spatial features. CNN is a deep learning model that automatically learns spatial dependency information in images. By training a CNN model on the preprocessed video frame data, the correlation between each pixel and surrounding pixels can be learned, thereby extracting facial feature data. These feature data can be used to further analyze the facial expression.
Long and short term memory networks (LSTM) are applied to analyze expression sequences. LSTM is a Recurrent Neural Network (RNN) that can capture long-term dependencies in time series data. By inputting the facial feature data into one LSTM model, the law of change of facial expression can be learned, and facial expression dynamic data can be generated. These data can help to better understand the trend of facial expression changes.
Facial feature data and facial expression dynamic data are integrated together, and facial expression recognition and emotion analysis reports are generated by adopting an emotion recognition algorithm. The emotion recognition algorithm can judge the emotion state expressed by the current facial expression according to the facial feature data and the facial expression dynamic data. By analyzing the emotional states of the plurality of samples, a detailed facial expression recognition and emotion analysis report can be generated, and a reference basis is provided for doctors.
Referring to fig. 7, based on the optimized depression treatment strategy and facial expression recognition and emotion analysis report, the clinical decision is adjusted in real time by adopting an online learning algorithm, and the steps of generating the optimized clinical decision and dynamic adjustment strategy are specifically as follows:
s601: based on the optimized depression treatment strategy, parameters are adjusted by using an online learning method, and initially adjusted clinical decision parameters are generated;
S602: optimizing decision parameters based on facial expression recognition and emotion analysis reports by combining reinforcement learning technology, and generating emotion-adjusted clinical decision parameters;
s603: based on the clinical decision parameters of emotion adjustment, real-time monitoring and dynamic adjustment are carried out, and self-adaptive control strategies are used for generating real-time monitoring adjustment data;
s604: based on the real-time monitoring adjustment data, final online learning optimization is executed, real-time performance and individuation of decisions are ensured, and optimized clinical decisions and dynamic adjustment strategies are generated.
And carrying out parameter adjustment on the optimized depression treatment strategy by using an online learning method. Online learning is a learning method that can continuously update model parameters in a data stream. By using facial expression recognition and emotion analysis reports as feedback signals, the therapeutic effect and emotion state of a patient can be monitored in real time, and parameters of a therapeutic strategy can be adjusted according to the information. This may make the treatment strategy more personalized and targeted.
Decision parameters are further optimized in conjunction with reinforcement learning techniques. Reinforcement learning is a method of learning optimal behavior through interactions with an environment. By reporting facial expression recognition and emotion analysis as reward signals, a reinforcement learning model can be trained to select the best clinical decision parameters. Thus, the treatment strategy can be more flexible and adaptive.
And carrying out real-time monitoring and dynamic adjustment, and generating real-time monitoring adjustment data by using an adaptive control strategy. Adaptive control is a method that can automatically adjust the control strategy according to the state change of the system. By monitoring the therapeutic effect and emotional state of the patient in real time, the parameters of the therapeutic strategy can be continuously adjusted to adapt to the changing requirements of the patient. This can improve the effect and satisfaction of the treatment.
And executing final online learning optimization, ensuring the real-time performance and individuation of the decision, and generating an optimized clinical decision and a dynamic adjustment strategy. By continuously and iteratively updating model parameters and using an adaptive control strategy, real-time optimization and personalized adjustment of clinical decisions can be realized. This can improve the effectiveness of the treatment and patient satisfaction.
Referring to fig. 8, an AI-based depression clinical decision system is used for executing the above-mentioned AI-based depression clinical decision method, and the system includes a data preprocessing module, a multi-mode feature extraction module, a feature fusion and mode recognition module, a treatment strategy optimization module, an emotion analysis module, a decision parameter optimization module, and a system integration and optimization module.
The data preprocessing module is used for carrying out frequency domain analysis by adopting a Fourier transform algorithm based on original multi-mode data of a depression patient, and processing a missing value by utilizing a mean value filling method and a nearest neighbor interpolation method to generate preprocessed multi-mode data;
the multi-modal feature extraction module performs complex sampling and time alignment by adopting a time sequence analysis method based on the preprocessed multi-modal data, automatically extracts key features by utilizing a Mel frequency cepstrum coefficient and a deep learning feature automatic encoder, and generates a time-synchronous multi-modal feature data set;
the feature fusion and pattern recognition module utilizes a time-synchronous multi-mode feature data set, utilizes a weighted fusion strategy to fuse the data, analyzes the patterns through a support vector machine algorithm, and generates fused multi-mode feature data and a psychological behavior pattern recognition result;
the treatment strategy optimization module sorts the importance of the features by adopting a random forest algorithm based on the psychological behavior pattern recognition result, and establishes a prediction model by utilizing a gradient lifting decision tree algorithm to generate baseline feature analysis data and treatment response prediction data;
the emotion analysis module extracts voice and video features by using a long-short-term memory network algorithm and a three-dimensional convolutional neural network according to the fused multi-modal feature data, analyzes the voice and video features by applying an emotion recognition algorithm, and generates voice video depth feature fusion data and facial expression recognition and emotion analysis reports;
The decision parameter tuning module is used for adjusting parameters based on treatment response prediction data and facial expression recognition and emotion analysis reports by utilizing an online learning method and a reinforcement learning technology to generate preliminarily adjusted clinical decision parameters and emotion adjusted clinical decision parameters;
the system integration and optimization module analyzes model performance by adopting a confusion matrix and a receiver operation characteristic curve, performs large-scale processing on data by a Z-score standardization method, trains a strategy network by utilizing a depth deterministic strategy gradient algorithm, performs Monte Carlo simulation in a simulation environment, collects decision expression, applies a depth deterministic strategy gradient algorithm refinement strategy again, and generates an optimized treatment strategy and an optimized depression treatment strategy.
The system can rapidly and accurately identify the psychological behavior mode and the emotional state of the patient through the analysis of the multimodal data and the automatic extraction of the key characteristics, thereby improving the diagnosis and treatment efficiency. Meanwhile, according to the feature importance ordering of the patient and the establishment of the prediction model, personalized treatment strategy optimization is realized. In addition, various information such as voice, video and facial expression are comprehensively considered, so that a more comprehensive diagnosis result is provided. The parameters are adjusted in real time through online learning and reinforcement learning technologies, and an optimized treatment strategy is generated by utilizing a depth deterministic strategy gradient algorithm, so that the treatment effect is further improved.
Referring to fig. 9, the data preprocessing module includes a frequency domain analysis sub-module, a missing value processing sub-module, and a data normalization sub-module;
the multi-mode feature extraction module comprises a time alignment sub-module, a key feature extraction sub-module and a feature data set construction sub-module;
the feature fusion and pattern recognition module comprises a data fusion sub-module, a pattern recognition sub-module and a feature cascade sub-module;
the treatment strategy optimization module comprises a feature importance sequencing sub-module, a prediction model construction sub-module and a model verification and tuning sub-module;
the emotion analysis module comprises a voice feature extraction sub-module, a video feature extraction sub-module and a feature fusion sub-module;
the decision parameter tuning module comprises a feature selection sub-module, a model selection sub-module, a super parameter tuning sub-module and a model evaluation sub-module;
the system integration and optimization module comprises a performance analysis sub-module, a data scale sub-module, a strategy network training sub-module, a Monte Carlo simulation sub-module and a strategy refinement sub-module.
In the data preprocessing module, a frequency domain analysis submodule adopts a Fourier transform algorithm to carry out frequency domain analysis on original multi-mode data of a patient suffering from depression; the missing value processing submodule processes missing values in the preprocessed multi-mode data by using a mean value filling method and a nearest neighbor interpolation method; and the data normalization sub-module performs normalization processing on the processed data.
In the multi-mode feature extraction module, a time alignment submodule carries out repeated sampling and time alignment on the preprocessed multi-mode data by adopting a time sequence analysis method; the key feature extraction submodule automatically extracts key features by utilizing the mel frequency cepstrum coefficient and the deep learning feature automatic encoder; the feature dataset construction submodule generates a time-synchronized multimodal feature dataset.
In the feature fusion and mode identification module, a data fusion submodule fuses the multi-mode feature data sets synchronized in time by using a weighted fusion strategy; the pattern recognition sub-module analyzes the fused characteristic data through a support vector machine algorithm to generate fused multi-pattern characteristic data; and the feature cascading sub-module cascades the fused multi-mode feature data with the psychological behavior mode recognition result.
In the treatment strategy optimization module, a feature importance ranking submodule adopts a random forest algorithm to rank the importance of features of the fused multi-mode feature data; the prediction model construction submodule establishes a prediction model by utilizing a gradient lifting decision tree algorithm; the model verification and tuning sub-module verifies and tunes the prediction model to generate baseline characteristic analysis data and treatment response prediction data.
In the emotion analysis module, a voice feature extraction sub-module extracts voice features by using a long-term memory network algorithm and a three-dimensional convolutional neural network; the video feature extraction submodule extracts video features by using a long-term and short-term memory network algorithm and a three-dimensional convolutional neural network; and the feature fusion sub-module fuses the voice features and the video features to generate voice video depth feature fusion data. And the facial expression recognition and emotion analysis report is generated by analyzing the fused multi-modal feature data and the voice video depth feature fusion data.
In the decision parameter tuning module, a feature selection submodule selects important features according to treatment response prediction data and facial expression recognition and emotion analysis reports; the model selection submodule selects a proper model for parameter adjustment; the super parameter tuning sub-module adjusts the super parameters of the model; the model evaluation sub-module evaluates the adjusted model performance.
In the system integration and optimization module, a performance analysis sub-module adopts a confusion matrix and a receiver operation characteristic curve to analyze the performance of the model; the data scale submodule carries out scale processing on the data by utilizing a Z score standardization method; the strategy network training submodule trains a strategy network by using a depth deterministic strategy gradient algorithm; the Monte Carlo simulation sub-module performs Monte Carlo simulation in a simulation environment and collects decision expressions; the strategy refinement sub-module applies the depth deterministic strategy gradient algorithm refinement strategy again, generating an optimized treatment strategy and an optimized depression treatment strategy.
The present invention is not limited to the above embodiments, and any equivalent embodiments which can be changed or modified by the technical disclosure described above can be applied to other fields, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matter of the present invention will still fall within the scope of the technical disclosure.
Claims (10)
1. An AI-based depression clinical decision method, comprising the steps of:
based on the depression patient, adopting a data preprocessing technology to clean and synchronize voice, video, text and physiological signals to generate multi-mode patient data;
based on the multi-modal patient data, adopting a long-short-term memory network and a three-dimensional convolutional neural network to analyze voice and video data, and carrying out feature fusion to generate a psychological behavior pattern recognition result;
based on the psychological behavior pattern recognition result, adopting a random forest and gradient lifting decision tree to predict the treatment response of the patient, evaluating the treatment effect, and generating a depression prediction result and a treatment effect evaluation report;
Based on the treatment effect evaluation report, adopting a depth deterministic strategy gradient algorithm to optimize a strategy, and testing on a simulation platform to generate an optimized depression treatment strategy;
based on video data in the multi-mode patient data, adopting a convolutional neural network and cyclic neural network mixed model to track facial expressions and analyze emotional states, and generating facial expression recognition and emotion analysis reports;
based on the optimized depression treatment strategy, facial expression recognition and emotion analysis report, adopting an online learning algorithm to adjust clinical decisions in real time, and generating optimized clinical decisions and dynamic adjustment strategies;
the multi-modal patient data is specifically a patient information set subjected to noise reduction, outlier processing and time stamp alignment, the psychological behavior pattern recognition result comprises an emotion state label, language expression habit and limb language characteristics, the depression prediction result is specifically probability distribution of future health states of the patient, the treatment effect evaluation report comprises benefit comparison of multiple treatment schemes, and the facial expression recognition and emotion analysis report comprises facial muscle activity data and emotion change-based time sequence analysis.
2. The AI-based depression clinical decision method of claim 1, wherein the step of generating multimodal patient data based on depression patient data by cleaning and synchronizing voice, video, text and physiological signals using data preprocessing techniques comprises:
performing frequency domain analysis by adopting Fourier transform based on original multi-mode data of a depression patient, and processing missing values by mean filling and nearest neighbor interpolation to generate preprocessed multi-mode data;
resampling and time alignment are carried out by adopting a time sequence analysis method based on the preprocessed multi-mode data, so as to generate time-synchronous multi-mode data;
based on the time-synchronous multi-mode data, extracting key features by adopting a Mel frequency cepstrum coefficient and a deep learning feature automatic encoder to generate a multi-mode feature data set;
and based on the multi-modal feature data set, adopting a weighted fusion strategy to fuse multi-modal data and generating fused multi-modal feature data.
3. The AI-based depression clinical decision method of claim 1, wherein based on the multimodal patient data, voice and video data analysis is performed using a long-term memory network and a three-dimensional convolutional neural network, and feature fusion is performed, and the step of generating a psychobehavioral pattern recognition result specifically comprises:
Based on the fused multi-mode characteristic data, extracting dynamic time characteristics of voice by adopting a long-short-period memory network, and generating voice time sequence characteristics;
based on the fused multi-mode characteristic data, extracting the spatial characteristics and dynamic changes of the video by adopting a three-dimensional convolutional neural network, and generating the space-time characteristics of the video;
based on the voice time sequence features and the video space-time features, performing depth feature fusion by using feature cascading and fusion strategies to generate voice video depth feature fusion data;
and carrying out pattern recognition by adopting a support vector machine based on the voice video depth feature fusion data to generate a psychological behavior pattern recognition result.
4. The AI-based depression clinical decision method of claim 1, wherein based on the psychographic pattern recognition result, the treatment response prediction of the patient is performed using a random forest and gradient boost decision tree, and the treatment effect is evaluated, and the step of generating a depression prediction result and a treatment effect evaluation report is specifically:
based on the psychological behavior pattern recognition result, ranking the feature importance by adopting a random forest algorithm, analyzing the change before and after treatment, and generating baseline feature analysis data;
Constructing a prediction model by adopting a gradient lifting decision tree based on the baseline characteristic analysis data, and performing treatment response prediction to generate treatment response prediction data;
based on the treatment response prediction data, performing model verification and tuning, and using k-fold cross verification to ensure generalization capability of the model to generate a treatment effect evaluation model;
and analyzing the model performance by adopting a confusion matrix and a receiver operation characteristic curve based on the treatment effect evaluation model, and generating a depression predicted result and a treatment effect evaluation report.
5. The AI-based depression clinical decision method of claim 1, wherein based on the treatment effect evaluation report, a depth deterministic strategy gradient algorithm optimization strategy is adopted, and the simulation platform is tested, and the step of generating an optimized depression treatment strategy specifically comprises:
based on the treatment effect evaluation report, performing scale processing on the data by adopting a Z-score standardization method, and generating normalized evaluation data;
based on the normalized evaluation data, training a strategy network by adopting a depth deterministic strategy gradient algorithm to generate a primarily optimized treatment strategy model;
Based on the primarily optimized treatment strategy model, performing Monte Carlo simulation in a simulation environment, collecting decision performance, and generating simulation test data;
and based on the simulation test data, the DDPG algorithm refinement strategy is applied again, and the optimized depression treatment strategy is generated.
6. The AI-based depression clinical decision method of claim 1, wherein tracking facial expressions and analyzing emotional states based on video data in the multimodal patient data using a convolutional neural network and cyclic neural network hybrid model, the step of generating facial expression recognition and emotional analysis reports is specifically:
based on video data in the multi-mode patient data, performing frame extraction and gray scale processing through an OpenCV tool to generate preprocessed video frame data;
based on the preprocessed video frame data, extracting spatial features by adopting a convolutional neural network to generate facial feature data;
based on the facial feature data, analyzing the expression sequence by applying a long-term memory network to generate facial expression dynamic data;
and based on the facial expression dynamic data, integrating facial feature data and facial expression dynamic data, and generating facial expression recognition and emotion analysis reports by adopting an emotion recognition algorithm.
7. The AI-based depression clinical decision method of claim 1, wherein based on the optimized depression treatment strategy and facial expression recognition and emotion analysis report, the clinical decision is adjusted in real time using an online learning algorithm, and the steps of generating the optimized clinical decision and dynamic adjustment strategy are specifically as follows:
based on the optimized depression treatment strategy, parameters are adjusted by using an online learning method, and initially adjusted clinical decision parameters are generated;
optimizing decision parameters based on the facial expression recognition and emotion analysis report in combination with reinforcement learning technology to generate emotion-adjusted clinical decision parameters;
based on the emotion-adjusted clinical decision parameters, real-time monitoring and dynamic adjustment are carried out, and self-adaptive control strategies are used for generating real-time monitoring adjustment data;
based on the real-time monitoring and adjusting data, final online learning optimization is executed, real-time performance and individuation of decisions are ensured, and optimized clinical decisions and dynamic adjusting strategies are generated.
8. An AI-based depression clinical decision system, characterized in that it comprises a data preprocessing module, a multi-modal feature extraction module, a feature fusion and mode recognition module, a treatment strategy optimization module, a mood analysis module, a decision parameter optimization module, and a system integration and optimization module according to the AI-based depression clinical decision method of any one of claims 1-7.
9. The AI-based depression clinical decision system of claim 8, wherein the data preprocessing module performs frequency domain analysis based on raw multi-modal data of a depression patient using a fourier transform algorithm and processes missing values using a mean-fill method and a nearest neighbor interpolation method to generate preprocessed multi-modal data;
the multi-modal feature extraction module performs complex sampling and time alignment by adopting a time sequence analysis method based on the preprocessed multi-modal data, automatically extracts key features by utilizing a Mel frequency cepstrum coefficient and a deep learning feature automatic encoder, and generates a time-synchronous multi-modal feature data set;
the feature fusion and pattern recognition module utilizes a time-synchronous multi-modal feature data set, utilizes a weighted fusion strategy to fuse data, analyzes patterns through a support vector machine algorithm, and generates fused multi-modal feature data and a psychological behavior pattern recognition result;
the treatment strategy optimization module sorts the importance of the features by adopting a random forest algorithm based on the psychological behavior pattern recognition result, and establishes a prediction model by utilizing a gradient lifting decision tree algorithm to generate baseline feature analysis data and treatment response prediction data;
The emotion analysis module extracts voice and video features by using a long-term memory network algorithm and a three-dimensional convolutional neural network according to the fused multi-modal feature data, analyzes the voice and video features by applying an emotion recognition algorithm, and generates voice video depth feature fusion data and facial expression recognition and emotion analysis reports;
the decision parameter tuning module is used for adjusting parameters based on treatment response prediction data and facial expression recognition and emotion analysis reports by utilizing an online learning method and a reinforcement learning technology to generate preliminarily adjusted clinical decision parameters and emotion adjusted clinical decision parameters;
the system integration and optimization module analyzes model performance by adopting a confusion matrix and a receiver operation characteristic curve, performs large-scale processing on data by a Z score standardization method, trains a strategy network by utilizing a depth deterministic strategy gradient algorithm, performs Monte Carlo simulation in a simulation environment, collects decision expression, applies a depth deterministic strategy gradient algorithm refinement strategy again, and generates an optimized treatment strategy and an optimized depression treatment strategy.
10. The AI-based depression clinical decision making system of claim 8, wherein the data preprocessing module comprises a frequency domain analysis sub-module, a missing value processing sub-module, a data normalization sub-module;
The multi-mode feature extraction module comprises a time alignment sub-module, a key feature extraction sub-module and a feature data set construction sub-module;
the feature fusion and pattern recognition module comprises a data fusion sub-module, a pattern recognition sub-module and a feature cascade sub-module;
the treatment strategy optimization module comprises a feature importance ordering sub-module, a prediction model construction sub-module and a model verification and tuning sub-module;
the emotion analysis module comprises a voice feature extraction sub-module, a video feature extraction sub-module and a feature fusion sub-module;
the decision parameter tuning module comprises a feature selection sub-module, a model selection sub-module, a super parameter tuning sub-module and a model evaluation sub-module;
the system integration and optimization module comprises a performance analysis sub-module, a data scale sub-module, a strategy network training sub-module, a Monte Carlo simulation sub-module and a strategy refinement sub-module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311664055.8A CN117747064A (en) | 2023-12-06 | 2023-12-06 | Depression clinical decision method and system based on AI |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311664055.8A CN117747064A (en) | 2023-12-06 | 2023-12-06 | Depression clinical decision method and system based on AI |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117747064A true CN117747064A (en) | 2024-03-22 |
Family
ID=90260063
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311664055.8A Pending CN117747064A (en) | 2023-12-06 | 2023-12-06 | Depression clinical decision method and system based on AI |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117747064A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117976148A (en) * | 2024-04-01 | 2024-05-03 | 成都市双流区妇幼保健院 | Machine learning-based prediction method and system for mental health problems of children |
-
2023
- 2023-12-06 CN CN202311664055.8A patent/CN117747064A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117976148A (en) * | 2024-04-01 | 2024-05-03 | 成都市双流区妇幼保健院 | Machine learning-based prediction method and system for mental health problems of children |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210201003A1 (en) | Synthetic data for neural network training using vectors | |
| CN108351862B (en) | Method and apparatus for determining developmental progress using artificial intelligence and user input | |
| JP2022009339A (en) | Platform and system for digital personalized medical treatment | |
| CN110188836B (en) | Brain function network classification method based on variational self-encoder | |
| CN117457147B (en) | Personalized nursing planning method and system for rehabilitation patient | |
| US11670421B2 (en) | Method and system enabling digital biomarker data integration and analysis for clinical treatment impact | |
| Vairachilai et al. | Body sensor 5 G networks utilising deep learning architectures for emotion detection based on EEG signal processing | |
| CN117438029A (en) | Intelligent evaluation system for wound severity of orthopedic wound patient | |
| CN117198468A (en) | Intervention scheme intelligent management system based on behavior recognition and data analysis | |
| Meshram et al. | Diagnosis of depression level using multimodal approaches using deep learning techniques with multiple selective features | |
| Bakariya et al. | Facial emotion recognition and music recommendation system using CNN-based deep learning techniques | |
| CN117009668A (en) | Self-adaptive learning content recommendation method and system based on deep reinforcement learning | |
| CN117747064A (en) | Depression clinical decision method and system based on AI | |
| CN113360772B (en) | Interpretable recommendation model training method and device | |
| Ahmadieh et al. | Visual image reconstruction based on EEG signals using a generative adversarial and deep fuzzy neural network | |
| CN106845386A (en) | A kind of action identification method based on dynamic time warping Yu Multiple Kernel Learning | |
| Kukkar et al. | A novel methodology using RNN+ LSTM+ ML for predicting student’s academic performance | |
| Wang et al. | Epileptic seizures prediction based on unsupervised learning for feature extraction | |
| Shaffi et al. | Multi-Planar MRI-Based Classification of Alzheimer's Disease Using Tree-Based Machine Learning Algorithms | |
| Eeswar et al. | Better you: Automated tool that evaluates mental health and provides guidance for university students | |
| CN117850601B (en) | System and method for automatically detecting vital signs of handheld PDA | |
| Moosavi et al. | Early mental stress detection using q-learning embedded starling murmuration optimiser-based deep learning model | |
| Fofanah et al. | Adaptive Neuro-Fuzzy Inference System with Non-Linear Regression Model for Online Learning Framework | |
| Bekenova et al. | Emotion recognition and classification based on audio data using AI | |
| CN117158904B (en) | Old people cognitive disorder detection system and method based on behavior analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |