Alzheimer disease early cognition evaluation system and method based on combination of traditional Chinese medicine and western medicine

Abstract

The invention relates to the technical field of medical evaluation, and discloses an Alzheimer's disease early cognition evaluation system and method based on combination of traditional Chinese medicine and western medicine. The system comprises a tongue image acquisition module, a pulse condition fluctuation signal acquisition module, a western medicine information acquisition module, a data preprocessing module, a feature extraction module, a data fusion module, a comprehensive data set establishment module and an evaluation decision module, wherein the tongue image acquisition module acquires tongue images, pulse condition fluctuation signals, inquiry text data and inquiry sound information; and the western medicine information acquisition module performs multi-modal fusion on feature data of traditional Chinese medicine and western medicine to construct a comprehensive data set. According to the system, multi-modal data is collected by integrating a traditional Chinese medicine four-diagnosis module and a western medicine detection module, the multi-modal data is processed and fused through algorithms including attention mechanism image processing, signal analysis and the like, and then an evaluation model is trained through integrated learning, transfer learning and the like. The method comprises the steps of equipment deployment, data acquisition and processing, model training, evaluation application and the like, multi-dimensional accurate evaluation is realized by combining optimization of a generative adversarial network, fuzzy logic and the like, and technical support is provided for early diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical evaluation, in particular to an Alzheimer's disease early cognitive evaluation system and method based on traditional Chinese and Western medicine combination. BACKGROUND

[0002] Alzheimer's disease is a serious threat to the health of the elderly neurodegenerative disease, its early diagnosis for delaying the development of the disease, improve patient outcomes has a key significance. Early cognitive evaluation as the core of diagnosis, need to analyze the patient's multi-dimensional physiological and pathological information, integration of traditional Chinese medicine holistic view and Western medicine precision detection advantages, the construction of multi-modal evaluation system has become an important direction of current research.

[0003] In the prior art, in the field of Alzheimer's disease early cognitive evaluation, TCM diagnosis mainly depends on the doctor to obtain patient information through four diagnostic methods of inspection, smell, ask and cut, and judge the disease type combined with experience; Western medicine mainly relies on neuroimaging examination (such as MRI, PET), neuropsychological scale test and blood biomarker detection method to evaluate cognitive function. Some research attempts to combine traditional Chinese and Western medicine diagnosis methods, but in the data processing level, traditional signal processing and image processing algorithms are mostly used, and the deep feature mining of traditional Chinese four diagnostic information (such as tongue texture and pulse waveform) is lacking; In terms of model construction, the model is mostly trained based on single modal data, which is difficult to realize the effective fusion of traditional Chinese and Western medicine data; In the evaluation decision-making link, the results are usually output by relying on fixed threshold or simple weighting method, which lacks dynamic adaptability.

[0004] However, the prior art still has many deficiencies. On the one hand, the collection and analysis of traditional Chinese four diagnostic information lack standardization and quantization means, resulting in strong subjectivity and low repeatability of the data; on the other hand, the traditional Chinese and Western medicine data fusion method is simple, and the complementary advantages of multi-modal data cannot be fully utilized, resulting in insufficient accuracy and robustness of the evaluation model. In addition, the existing evaluation system mostly focuses on single dimension index, which is difficult to meet the clinical demand for multi-dimensional, personalized and accurate evaluation. In view of this, we propose an Alzheimer's disease early cognitive evaluation system and method based on traditional Chinese and Western medicine combination. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides an Alzheimer's disease early cognitive evaluation system and method based on traditional Chinese and Western medicine combination, which solves the problem that single dimension index is difficult to meet the clinical demand for multi-dimensional, personalized and accurate evaluation.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: an Alzheimer's disease early cognitive evaluation system and method based on traditional Chinese and Western medicine combination, comprising the following modules:

[0007] The TCM information collection module is configured to collect tongue image, pulse wave signal, inquiry text data and smell diagnosis sound information;

[0008] The western medicine information collection module is configured to connect a medical imaging device to obtain brain image, collect cognitive function test scale data, and obtain blood biochemical detection index;

[0009] The data preprocessing module is configured to perform noise reduction on the collected image, signal filtering and text standardization processing.

[0010] The feature extraction module is configured to extract tongue features by image processing algorithm, analyze pulse features by signal analysis algorithm, extract key information of inquiry by natural language processing technology, and extract relevant features from western medicine image and detection data.

[0011] The data fusion module is configured to perform multi-modal fusion on the feature data of TCM and western medicine, and construct a comprehensive data set.

[0012] The evaluation decision module is configured to train and model the fusion data set based on deep learning algorithm, output cognitive evaluation result and transfer to the feedback unit.

[0013] Preferably, the data processing unit adopts an image processing algorithm based on attention mechanism to segment and extract features of the tongue image. Specifically, by constructing a spatial attention module and a channel attention module, the spatial position and feature channel of the tongue image are respectively weighted, and the texture and color features of the key region of the tongue body are highlighted. In the segmentation process, the level set algorithm is combined with the active contour model to automatically identify the edge of the tongue body and segment the tongue body region.

[0014] Preferably, the data processing unit uses wavelet transform algorithm to filter the pulse waveform data. First, the pulse signal is decomposed into different frequency subbands to remove high-frequency noise and baseline drift. Then, the frequency, amplitude and morphological multi-scale feature parameters are extracted by combining the empirical mode decomposition algorithm. During the empirical mode decomposition, the intrinsic mode function decomposition termination condition is judged by screening the stopping criterion. At the same time, the Hilbert-Huang transform is used to analyze the time-frequency characteristics of the decomposed intrinsic mode function.

[0015] Preferably, the data processing unit uses a natural language processing algorithm based on bidirectional long short-term memory network combined with attention mechanism to perform semantic understanding and key symptom information extraction on the inquiry and smell diagnosis information. A multi-layer Bi-LSTM network structure is constructed to capture the context semantic information of the text. The multi-head attention mechanism is introduced to calculate the correlation degree between words in the text sequence from different angles, and the extraction ability of key symptom information is enhanced.

[0016] Preferably, the evaluation unit adopts an ensemble learning algorithm to fuse multiple deep learning models of different structures, and in the fusion process, the weight of each sub-model is determined through cross-validation. For convolutional neural networks, the receptive field of the network is expanded by using residual connection structure and dilated convolution technology, and finally the final cognitive evaluation result is output through weighted voting or stacked generalization.

[0017] Preferably, the evaluation unit is based on a transfer learning algorithm, uses pre-trained model parameters on a large-scale medical image dataset, and fine-tunes the model with Alzheimer's disease-related traditional Chinese and Western medicine combined data. The pre-trained DenseNet network is selected as the base model, and in the fine-tuning process, a progressive thawing strategy is adopted, that is, the bottom layer parameters of the network are fixed first, only the top layer parameters are trained, and the bottom layer parameters are gradually thawed as the training progresses.

[0018] The method comprises the following steps:

[0019] S1: device deployment step

[0020] Purchase special collection equipment for traditional Chinese four diagnostic methods and complete debugging, and establish data interface connection of Western medicine detection equipment;

[0021] S2: data acquisition step

[0022] Respectively use the traditional Chinese four diagnostic information collection module and the Western medicine detection information collection module to obtain traditional Chinese four diagnostic data and Western medicine detection data of the patient;

[0023] S3: data cleaning step

[0024] The collected data is processed for deduplication, missing data filling and format unification;

[0025] S4: feature mining step

[0026] Use image processing, signal analysis, natural language processing and other technologies to extract traditional Chinese medicine data features, and use medical image analysis and biochemical index analysis methods to extract Western medicine data features;

[0027] S5: model training step

[0028] After fusing the extracted traditional Chinese and Western medicine feature data, an early cognitive evaluation model for Alzheimer's disease is trained based on a deep learning algorithm;

[0029] S6: evaluation execution step

[0030] The processed new collected data is input into the trained model to obtain and output the cognitive evaluation result.

[0031] Preferably, in the data processing step, an image enhancement algorithm based on a generative adversarial network is used to preprocess the tongue image. A GAN network structure containing a generator and a discriminator is constructed. The generator adopts a U-Net architecture, which preserves the detailed information of the image through a skip connection. The discriminator adopts a PatchGAN structure to judge the authenticity of the local area of the image.

[0032] Preferably, in the training process, a gradient penalty mechanism is introduced to ensure the stability of the training of the generator and the discriminator. Through the adversarial training of the generator and the discriminator, the clarity and feature recognition of the tongue image are improved, and the enhanced image is further optimized by histogram equalization and adaptive histogram equalization.

[0033] Preferably, in the evaluation application step, a decision algorithm based on fuzzy logic is used to process the results output by the evaluation model. A fuzzy logic system containing input variables, a fuzzy rule base and output variables is constructed. The input variables are multiple dimension evaluation indexes output by the evaluation model, including the probability of disease and the cognitive function score. The fuzzy rule base is determined by expert experience and historical data. The input variables are mapped to different fuzzy sets, and the fuzzy output is converted to clear numerical values. Considering multiple dimension evaluation indexes, the weights of each index are determined by the analytic hierarchy process, and finally the final cognitive evaluation conclusion is generated.

[0034] The application provides an early cognitive evaluation system and method for Alzheimer's disease based on the combination of traditional Chinese medicine and Western medicine.

[0035] The application has the following advantages:

[0036] 1. The application integrates traditional Chinese medicine four examinations and Western medicine detection modules to collect multi-modal data. The data are processed and fused by image processing and signal analysis algorithms containing attention mechanisms, and then an evaluation model is trained by using ensemble learning and transfer learning. The method covers device deployment, data collection and processing, model training and evaluation application, and realizes multi-dimensional accurate evaluation by combining generative adversarial networks and fuzzy logic, thereby providing technical support for early diagnosis.

[0037] 2. The application uses a bidirectional long short-term memory network (Bi-LSTM) combined with an attention mechanism to process inquiries and inquiry information. Bi-LSTM captures the semantic dependence of text context through bidirectional propagation, and abstracts features layer by layer through a multi-layer structure. The multi-head attention mechanism calculates the word association weight from multiple dimensions and focuses on key symptom words. After processing by the embedding layer, Bi-LSTM layer, attention layer and classification layer, the symptom information can be accurately extracted. This method enhances the semantic understanding and key information extraction ability, improves the generalization and robustness of the model, and provides effective support for the digitization and intelligent auxiliary diagnosis of traditional Chinese medicine four examinations information.

[0038] 3、The application adopts the mode of combining fuzzy logic decision algorithm with analytic hierarchy process (AHP), the fuzzy logic fuzzes multi-dimensional indexes such as disease probability and cognitive function score, rules library reasoning is constructed through expert experience and historical data, and clear values are obtained through defuzzification; AHP determines the weight of each index, and generates evaluation conclusion by integrating multi-dimensional results. This method effectively handles the uncertainty in evaluation, realizes scientific integration of multiple indexes, improves accuracy and interpretability, and can flexibly adapt to different scene requirements, providing reliable support for cognitive evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The Alzheimer's disease early cognitive evaluation system based on the combination of traditional Chinese and Western medicine in the application is a flow chart;

[0040] Figure 2 The image processing algorithm flowchart based on the attention mechanism in the application is a flowchart;

[0041] Figure 3 The Alzheimer's disease early cognitive evaluation method based on the combination of traditional Chinese and Western medicine in the application is a flow chart;

[0042] Figure 4 The decision algorithm flowchart based on fuzzy logic in the application is a flowchart. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the application will be described clearly and completely in the drawings in the specification of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0044] Embodiment:

[0045] Please refer to the drawings in the specification of the application Figure 1 -Appendix Figure 2 The Alzheimer's disease early cognitive evaluation system based on the combination of traditional Chinese and Western medicine in the embodiments of the application comprises the following modules:

[0046] The traditional Chinese medicine information acquisition module is configured to acquire tongue image, pulse wave signal, inquiry text data and smell diagnosis sound information;

[0047] The Western medicine information acquisition module is set to connect medical imaging equipment to obtain brain image, acquire cognitive function test scale data, and obtain blood biochemical detection indexes;

[0048] The data preprocessing module is used for denoising the acquired image, filtering the signal, and standardizing the text;

[0049] The feature extraction module extracts tongue image features through image processing algorithms, analyzes pulse features using signal analysis algorithms, extracts key information from the inquiry using natural language processing technology, and extracts relevant features from western medical images and test data;

[0050] The data fusion module fuses the feature data of traditional Chinese medicine and western medicine in multiple modalities to construct a comprehensive data set;

[0051] The evaluation decision module trains the fusion data set based on a deep learning algorithm, outputs the cognitive evaluation result, and transmits it to the feedback unit.

[0052] The data processing unit uses an image processing algorithm based on an attention mechanism to segment and extract features from tongue image, specifically by constructing a spatial attention module and a channel attention module to assign weights to the spatial position and feature channels of the tongue image, highlighting the texture and color features of the key area of the tongue. In the segmentation process, the level set algorithm is used in combination with the active contour model to automatically identify the tongue edge and segment the tongue area. The following algorithm is established here:

[0053] Step 1: Image input and enhancement preprocessing

[0054] After obtaining the original tongue image I, non-local mean filtering combined with guided filtering is used for preprocessing. Non-local mean filtering uses the weighted average of similar blocks in the image to suppress noise while preserving texture details; guided filtering uses the original image as a guide to smooth the edges of the filtered results, further enhancing the contrast between the tongue and the background, and providing high-quality images for subsequent feature extraction;

[0055] 1. Non-local mean filtering:

[0056]

[0057] where Ω(i) is a search window centered on pixel i, w(i,j) is the weight between pixels i and j, calculated according to the similarity of pixel blocks, and C(i) is a normalization constant to ensure that the brightness of the filtered image remains unchanged;

[0058] 2. Guided filtering:

[0059]

[0060] where q(i) is the output pixel value, ωi is a local window centered on pixel i, Ij is the pixel value of the guide image (here, the original image I) in the window, and are the mean values of the guide image and the image to be filtered in the window, respectively;

[0061] Step 2: Construct a dynamic spatio-temporal and channel semantic joint attention module

[0062] The pre-processed feature map First, the spatial and channel dimension feature information is captured by the spatio-temporal convolution layer; then, the dynamic routing algorithm is used to dynamically allocate the spatio-temporal attention weight, and focus on the key area adaptively; then, the multi-head self-attention mechanism and the feedforward neural network of the Transformer architecture are introduced to mine the long-distance dependence and semantic association between channels, and finally the enhanced feature map is output

[0063] 1. Spatio-temporal convolution:

[0064] F conv =Conv ST (F)

[0065] Where Conv ST represents the spatio-temporal convolution operation, which extracts spatio-temporal features by a specific convolution kernel in the spatial and channel dimensions;

[0066] 2. Dynamic routing:

[0067]

[0068] b ij =tanh(u j|i )

[0069]

[0070] Where u j|i represents the predicted vector from capsule i to capsule j, w ijk is the connection weight, s ik is the input vector of capsule i, b ij is the routing logarithm, s j is the output vector of capsule j, and the routing logarithm is updated by iteration to realize the dynamic allocation of spatio-temporal attention weight;

[0071] 3. Multi-head self-attention:

[0072]

[0073] MultiHead(Q,K,V)=Concat(head1,...,head h )W O

[0074] Where Q, K, and V are query, key, and value matrices, respectively, d k is the dimension of the key, head i is the output of the i-th attention head, W O is the output weight matrix, and the relationship between channels is captured from different angles by the multi-head mechanism;

[0075] 4. Feedforward neural network:

[0076] FFN(x) = ReLU(xW1 + b1)W2 + b2

[0077] where x is the input feature, W1, W2 and b1, b2 are the weights and biases of the two layers of neural network, respectively, and the attention output is nonlinearly transformed.

[0078] Step 3: Attention feature fusion and deep learning initialization

[0079] First, the feature maps output by the joint attention module are fused with the original features through a fusion strategy based on the gating mechanism combined with residual connection; then the fused feature maps are used to perform semantic segmentation using a pre-trained U-Net++ network to generate a tongue region probability map, which is converted into a directed distance field to initialize the level set function φ(x, y, 0).

[0080] Formula representation:

[0081] Gating mechanism fusion:

[0082] g = σ(W g [F DTSCSAM ; F] + b g )

[0083] F fusion = g o F DTSCSAM + (1 - g) o F

[0084] where g is the gating vector, σ is the Sigmoid function, W g and b g are the weights and biases of the gating unit, o represents element multiplication, F DTSCSAM is the output of the joint attention module, and F is the original input feature map.

[0085] Residual connection:

[0086] F enhanced = F fusion + F

[0087] U-Net++ semantic segmentation and initialization:

[0088] P = U-Net++(F enhanced )

[0089]

[0090] where P is the probability map of the tongue region, and the initial state of the level set function is constructed through directed distance field conversion.

[0091] Step 4: Self-adaptive parameter level set-GAN collaborative evolution

[0092] During the evolution of the level set function, the parameters of the evolution equation are dynamically adjusted according to the matching degree of the contour and the image features; meanwhile, a generative adversarial network is introduced, the generator generates a predicted contour based on the current contour and image features, the discriminator distinguishes between true and false, and the adversarial training loss is integrated into the active contour energy function optimization to realize the collaborative evolution of the two:

[0093] 1. Adaptive coefficient calculation:

[0094]

[0095]

[0096] Where, ν(t) and λ(t) are the expansion / contraction term coefficient and edge indicator function weight changing with time t respectively, ν0 and λ0 are the initial coefficients, T is the total iteration number, dist(C t ,Edge) is the distance between the current contour C t and the target edge, and maxDist is the maximum distance threshold;

[0097] 2. Level set evolution:

[0098]

[0099] 3. Generative adversarial network:

[0100] Generator:

[0101] G(x)=ConvTranspose(FC(Concat(x,F enhanced ))

[0102] Discriminator:

[0103] D(y)=softmax(FC(Conv(y)))

[0104] Adversarial training loss:

[0105] L GAN =Ey real [logD(y real )]+E x [log(1-D(G(x)))]

[0106] Integrate L GAN into the active contour energy function optimization.

[0107] Step 5: Contour extraction and multi-scale feature initial extraction

[0108] When the level set function evolves stably, the zero contour line is extracted as the tongue contour. The color histogram features and local binary pattern (LBP) texture features of the tongue region are preliminarily extracted at different resolutions to prepare for subsequent deep feature fusion.

[0109] The formula is:

[0110] Color histogram:

[0111] H c = histogram(I tongue )

[0112] LBP texture features:

[0113]

[0114] where g c is the gray value of the center pixel, g i is the gray value of the neighborhood pixel, and s(x) is the sign function.

[0115] Step 6: Structure feature extraction based on graph convolution network

[0116] The tongue region is constructed as a graph structure, with nodes representing local regions of the image and edges representing the relationship between regions. The graph convolution network (GCN) is used to process the graph structure data to extract a feature matrix X GCN that reflects the structure information of the tongue, and to mine the internal structure correlation of the tongue. The formula is:

[0117] X GCN = GCN(X graph )

[0118] where X graph is the node feature matrix after the tongue region is constructed as a graph structure, and the structure feature X

[0119] Step 7: Multi-scale feature fusion and output

[0120] The feature pyramid network (FPN) is used to fuse the color and texture features extracted in step 5 and the structure features extracted in step 6. Through upsampling, downsampling and fusion operations, different scales and types of feature information are integrated to output a comprehensive feature vector containing rich details and semantics for Alzheimer's disease cognitive assessment.

[0121] The formula is:

[0122] F final = FPN(H c , LBP, X GCN )

[0123] The data processing unit filters the pulse wave data by using a wavelet transform algorithm, decomposes the pulse signal into different frequency subbands first, removes high-frequency noise and baseline drift, extracts multi-scale feature parameters including frequency, amplitude and morphology in the pulse wave by combining an empirical mode decomposition algorithm, judges the decomposition termination condition of the intrinsic mode function by screening the stopping criterion during empirical mode decomposition, and performs time-frequency analysis on the decomposed intrinsic mode function by using a Hilbert-Huang transform to obtain the time-frequency characteristics of the pulse signal.

[0124] The data processing unit uses a natural language processing algorithm of a bidirectional long short-term memory network combined with an attention mechanism to perform semantic understanding and key symptom information extraction on the inquiry and auscultation information, constructs a multi-layer Bi-LSTM network structure to capture the context semantic information of the text, introduces a multi-head attention mechanism to calculate the correlation degree between words in the text sequence from different angles, and enhances the extraction ability of key symptom information. The following algorithm is established here:

[0125] Step 1: Data preprocessing and embedding layer construction

[0126] Convert the inquiry and inquiry text into a vector representation that can be processed by the model. First, perform word segmentation, construct a vocabulary table, and map each word to an index; then convert the word index to a low-dimensional dense vector through an embedding layer.

[0127] Formula representation:

[0128] Word index mapping: w i →index(w i )(i=1,2,…,n)

[0129] Embedding layer conversion: Where w i is the i-th word, and d is the embedding dimension.

[0130] Step 2: Construction of bidirectional LSTM layer

[0131] Use bidirectional LSTM to capture the context information of the text sequence. The forward LSTM processes the sequence from left to right, and the backward LSTM processes the sequence from right to left. Finally, the hidden states of the two directions are spliced.

[0132] Formula representation:

[0133] Forward LSTM hidden state:

[0134] Backward LSTM hidden state:

[0135] Bidirectional splicing: Where h is the hidden layer dimension of the unidirectional LSTM

[0136] Step 3: Multi-head attention mechanism implementation

[0137] Through the multi-head attention mechanism, the correlation between words in the text sequence is calculated from different angles, enhancing the extraction ability of key symptom information.

[0138] Formula representation:

[0139] Linear projection generates query, key, value matrix: Q = W Q H, K = W K H, V = W V H, where H = [h1, h2, …, h n ] is the Bi-LSTM output sequence, W Q , W K , W V is the projection matrix.

[0140] Scaled dot-product attention calculation:

[0141] Multi-head parallel calculation and integration:

[0142] MultiHead(Q, K, V) = Concat(head1, …, head h )W O

[0143] where, h is the number of heads

[0144] Step 4: Attention weight application and context vector generation

[0145] According to the attention weight, the output of Bi-LSTM is weighted and summed to generate a context vector containing key symptom information.

[0146] Formula representation:

[0147] Attention weight calculation: α t = softmax(W a ·tanh(W h ·h t +b h )+b a ) where W a , W h are weight matrices, and b a , b h are bias terms.

[0148] Context vector generation:

[0149] Step 5: Key symptom extraction and classification

[0150] Based on the context vector, the symptom classification or label prediction is performed through a fully connected layer to extract key symptom information from the text.

[0151] The formula is:

[0152] Symptom classification: Where, is the predicted symptom label probability distribution, W c is the classification weight matrix, and b c is the bias term

[0153] Step 6: Model training and optimization

[0154] The model is trained using labeled data, and the model parameters are optimized by minimizing the loss function.

[0155] The formula is:

[0156] The cross-entropy loss function is: Where N is the number of samples, C is the number of classes, y ij is the true label, is the predicted probability.

[0157] Parameter update:

[0158] Where θ is the model parameter, and η is the learning rate.

[0159] The evaluation unit adopts an ensemble learning algorithm to fuse multiple deep learning models with different structures. In the fusion process, the weights of each sub-model are determined through cross-validation. For convolutional neural networks, residual connection structures and dilated convolution techniques are used to expand the network's receptive field. Finally, the final cognitive evaluation result is output through weighted voting or stacked generalization.

[0160] The evaluation unit is based on a transfer learning algorithm. It uses pre-trained model parameters on a large-scale medical image dataset and fine-tunes them with Alzheimer's disease-related integrated traditional Chinese and Western medicine data. The pre-trained DenseNet network is selected as the base model. During fine-tuning, a progressive unfreezing strategy is used. The bottom layer parameters of the network are fixed first, and only the top layer parameters are trained. As the training progresses, the bottom layer parameters are gradually unfrozen.

[0161] Please refer to the attached Figure 3 - attached Figure 4 The early cognitive evaluation method for Alzheimer's disease based on integrated traditional Chinese and Western medicine includes the following steps:

[0162] S1: Device deployment step

[0163] Purchase and debug the four diagnostic collection equipment for traditional Chinese medicine, and establish the data interface connection of the Western medicine detection equipment;

[0164] S2: data collection step

[0165] Respectively apply the traditional Chinese medicine four diagnostic information collection module and the western medicine detection information collection module to obtain traditional Chinese medicine four diagnostic data and western medicine detection data of the patient;

[0166] S3: data cleaning step

[0167] The collected data is processed for deduplication, missing data filling and format unification;

[0168] S4: feature mining step

[0169] Image processing, signal analysis, natural language processing and other technologies are used to extract traditional Chinese medicine data features, and medical image analysis and biochemical index analysis methods are used to extract western medicine data features;

[0170] S5: model training step

[0171] After the extracted traditional Chinese and western medicine feature data is fused, an early cognitive assessment model for Alzheimer's disease is trained based on a deep learning algorithm;

[0172] S6: evaluation execution step

[0173] The new collected data is input into the trained model after processing, and the cognitive assessment result is obtained and output.

[0174] In the data processing step, an image enhancement algorithm based on a generative adversarial network is used to preprocess the tongue image. A GAN network structure containing a generator and a discriminator is constructed. The generator uses a U-Net architecture to retain the detail information of the image through a skip connection; the discriminator uses a PatchGAN structure to judge the authenticity of the local area of the image.

[0175] In the training process, a gradient penalty mechanism is introduced to ensure the stability of the training of the generator and the discriminator; through the adversarial training of the generator and the discriminator, the clarity and feature recognition of the tongue image are improved, and the enhanced image is further optimized by histogram equalization and adaptive histogram equalization.

[0176] In the evaluation application step, a decision algorithm based on fuzzy logic is used to perform secondary processing on the results output by the evaluation model, a fuzzy logic system containing input variables, a fuzzy rule base and output variables is constructed, the input variables are multiple dimension evaluation indexes output by the evaluation model, including the probability of disease and the cognitive function score, the fuzzy rule base is determined through expert experience and historical data, the input variables are mapped to different fuzzy sets, and the fuzzy output is converted to clear numerical values; considering the multi-dimensional evaluation indexes, the weights of each index are determined by combining the analytic hierarchy process, and finally the final cognitive evaluation conclusion is generated; the following algorithm is established for the evaluation process:

[0177] Step 1: Define input variables and fuzzy sets

[0178] Multiple dimensional indicators output by the assessment model (e.g. disease probability, cognitive function score) are taken as input variables of the fuzzy logic system, and multiple fuzzy sets (e.g. "low", "medium", "high") are defined for each variable.

[0179] Assume disease probability P (range: 0-1):

[0180]

[0181] Cognitive function score S (range: 0-100): Similarly define membership functions for fuzzy sets "low", "medium", "high"

[0182] Step 2: Construct fuzzy rule base

[0183] Based on expert experience and historical data, define logical relationships between input variables and output variable (risk of cognitive impairment R).

[0184] Rule 1: IF P is "high" AND S is "low" THEN R is "high";

[0185] Rule 2: IF P is "medium" AND S is "medium" THEN R is "medium";

[0186] Rule 3: IF P is "low" AND S is "high" THEN R is "low";

[0187] General form: IF x1 is A i AND x2 is B j THEN y is C i jwhere x1, x2 are input variables, A i , B j are fuzzy sets, y is output variable, C ij is output fuzzy set;

[0188] Step 3: Fuzzy inference (calculate rule activation degree)

[0189] Text explanation: For each rule, calculate its activation degree (i.e. the degree of satisfaction of the premise) according to the current input values.

[0190] Formula representation:

[0191] Mamdani minimization method:

[0192] α1 = min(μ 高 (P), μ 低 (S))

[0193] a2 = min(μ 中 (P), μ 中 (S))

[0194] a3 = min(μ 低 (P), μ 高 (S))

[0195] Step 4: Aggregation of Rules Output

[0196] Textual Explanation: Combine the output of all rules to form the final fuzzy output set.

[0197] Formula Representation:

[0198] Maximum Aggregation: μ R (r) = max(a1 · μ 高 (r), a2 · μ 中 (r), a3 · μ 低 (r)) where μ R (r) is the final membership function for risk R

[0199] Step 5: Defuzzification (Conversion to Crisp Value)

[0200] Convert the fuzzy output set to a specific numerical result.

[0201] Center of Gravity (COG):

[0202] Discrete Form (if r is a discrete value):

[0203] Step 6: Constructing Analytic Hierarchy Process (AHP) Judgment Matrix

[0204] Determine the relative importance of each evaluation index (e.g., probability of illness, cognitive score) through expert evaluation.

[0205] Formula Representation:

[0206] Judgment Matrix A (Assuming two indices P and S):

[0207] where a 12 = 3 indicates that index P is slightly more important than S

[0208] Step 7: Calculating Weight Vector and Consistency Check

[0209] Extract the weight vector from the judgment matrix and verify the consistency of the judgment.

[0210] Formula Representation:

[0211] Eigenvalue method to find weight: A·W = λ max ·W

[0212] Solve the largest eigenvalue λ max The corresponding eigenvector W, after normalization, get weight vector w = [w1, w2].

[0213] Consistency check:

[0214] Where, n = 2, RI = 0 (need special processing, usually n ≥ 3 test is meaningful).

[0215] Step 8: comprehensive evaluation results

[0216] Combined with fuzzy logic output and AHP weight, generate the final evaluation conclusion;

[0217] Formula:

[0218] Weighted final risk value: FinalRisk = w1·R final +w2·S final

[0219] Where, S final is the final value of another dimension evaluation.

[0220] Risk level mapping:

[0221] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. The early cognitive assessment system for Alzheimer's disease based on the integration of traditional Chinese and Western medicine is characterized by: Includes the following modules: A traditional Chinese medicine information collection module configured to collect tongue images, pulse fluctuation signals, medical consultation text data, and auscultation sound information; A Western medicine information acquisition module is configured to connect to medical imaging equipment to obtain brain images, collect cognitive function test scale data, and obtain blood biochemical test indicators; Data preprocessing module, used to reduce noise on collected images, filter signals, and standardize text; The feature extraction module uses image processing algorithms to extract tongue features, signal analysis algorithms to analyze pulse features, and natural language processing technology to extract key information from medical interviews. It also extracts relevant features from Western medical imaging and test data. The data fusion module integrates the characteristic data of traditional Chinese medicine and Western medicine into a multimodal fusion to construct a comprehensive data set; The evaluation and decision-making module trains and models the fusion data set based on the deep learning algorithm, outputs the cognitive evaluation results and transmits them to the feedback unit.

2. The early cognitive assessment system for Alzheimer's disease based on the integration of traditional Chinese and Western medicine according to claim 1 is characterized in that: The data processing unit uses an image processing algorithm based on the attention mechanism to segment and extract features from the tongue image. Specifically, by constructing a spatial attention module and a channel attention module, weights are assigned to the spatial position and feature channels of the tongue image respectively, highlighting the texture and color features of the key areas of the tongue. During the segmentation process, the level set algorithm is combined with the active contour model to automatically identify the tongue edge and segment the tongue area.

3. The early cognitive assessment system for Alzheimer's disease based on the integration of traditional Chinese and Western medicine according to claim 1 is characterized in that: The data processing unit utilizes a wavelet transform algorithm to filter the pulse waveform data, first decomposes the pulse signal into different frequency sub-bands, removes high-frequency noise and baseline drift, and extracts multi-scale characteristic parameters including frequency, amplitude, and morphology from the pulse signal in combination with an empirical mode decomposition algorithm. During the empirical mode decomposition, the decomposition termination condition of the intrinsic mode function is judged by screening a stopping criterion, and simultaneously utilizes a Hilbert-Huang transform to perform time-frequency analysis on the decomposed intrinsic mode function to obtain the time-frequency characteristics of the pulse signal.

4. The Alzheimer's disease early cognitive assessment system based on the integration of traditional Chinese and Western medicine according to claim 1 is characterized in that: The data processing unit uses a bidirectional long short-term memory network combined with a natural language processing algorithm of an attention mechanism to perform semantic understanding of medical consultation and auscultation information and extract key symptom information, and constructs a multi-layer Bi-LSTM network structure to capture the contextual semantic information of the text. It introduces a multi-head attention mechanism to calculate the degree of correlation between words in a text sequence from different angles, thereby enhancing the ability to extract key symptom information.

5. The early cognitive assessment system for Alzheimer's disease based on the integration of traditional Chinese and Western medicine according to claim 1 is characterized in that: The evaluation unit adopts an integrated learning algorithm to fuse multiple deep learning models with different structures. During the fusion process, the weight of each sub-model is determined through cross-validation. For convolutional neural networks, residual connection structure and void convolution technology are used to expand the network's receptive field, and finally the final cognitive evaluation result is output through weighted voting or stacked generalization.

6. The Alzheimer's disease early cognitive assessment system based on the integration of traditional Chinese and Western medicine according to claim 1 is characterized in that: The evaluation unit is based on a transfer learning algorithm, using model parameters pre-trained on a large-scale medical imaging dataset, combined with traditional Chinese and Western medicine data related to Alzheimer's disease for fine-tuning, selecting a pre-trained DenseNet network as the basic model, and adopting a progressive unfreezing strategy during the fine-tuning process, first fixing the underlying parameters of the network and only training the top-level parameters, and gradually unfreezing the underlying parameters as training progresses.

7. The method for early cognitive assessment of Alzheimer's disease based on integration of traditional Chinese and Western medicine according to claim 1, characterized in that: The following steps are involved: S1: Device deployment steps Purchase and debug dedicated data acquisition equipment for the four diagnostic methods of Traditional Chinese Medicine, and establish data interface connections for Western medicine testing equipment; S2: Data collection step Use the TCM Four Diagnosis Information Collection Module and the Western Medicine Test Information Collection Module respectively to obtain the patient's TCM Four Diagnosis Data and Western Medicine Test Data; S3: Data cleaning step De-duplicate, fill in gaps and unify the format of collected data; S4: Feature mining step Image processing, signal analysis, natural language processing and other technologies are used to extract the characteristics of traditional Chinese medicine data, and medical image analysis and biochemical index analysis methods are used to extract the characteristics of Western medicine data; S5: Model training step After fusing the extracted traditional Chinese and Western medicine feature data, an early cognitive assessment model for Alzheimer's disease was trained based on a deep learning algorithm; S6: Evaluation execution steps The newly collected data is processed and input into the trained model to obtain and output the cognitive assessment results.

8. The system and method for early cognitive assessment of Alzheimer's disease based on integration of traditional Chinese and Western medicine according to claim 7, characterized in that: During the data processing step, tongue images were preprocessed using an image enhancement algorithm based on a generative adversarial network. A GAN network structure was constructed, comprising a generator and a discriminator. The generator used a U-Net architecture, preserving image details through skip connections; the discriminator employed a PatchGAN structure to determine the authenticity of local regions in the image.

9. The system and method for early cognitive assessment of Alzheimer's disease based on integration of traditional Chinese and Western medicine according to claim 8, characterized in that: During the training process, a gradient penalty mechanism is introduced to ensure the training stability of the generator and the discriminator; through adversarial training of the generator and the discriminator, the clarity and feature recognition of the tongue image are improved, and histogram equalization and adaptive histogram equalization are used to further optimize the enhanced image.

10. The system and method for early cognitive assessment of Alzheimer's disease based on integration of traditional Chinese and Western medicine according to claim 7, characterized in that: In the evaluation application step, a decision algorithm based on fuzzy logic is used to perform secondary processing on the results output by the evaluation model, and a fuzzy logic system is constructed, which includes input variables, a fuzzy rule base, and output variables. The input variables are multiple dimensional evaluation indicators output by the evaluation model, including disease probability and cognitive function score. The fuzzy rule base is determined through expert experience and historical data, and the input variables are mapped to different fuzzy sets, and the fuzzy output is converted into a clear numerical value. Comprehensively consider the multi-dimensional evaluation indicators, combine the hierarchical analysis method to determine the weight of each indicator, and finally generate the final cognitive evaluation conclusion.

Images