Diagnostic system and method

Abstract

The invention discloses a diagnosis system and method, and the system comprises a data collection module which is configured to obtain a Raman spectrum image of a target tissue; the image processing module is configured to pre-process the Raman spectrum image to obtain a pre-processed Raman spectrum image; the feature acquisition module is configured to extract spectral features and image features in the preprocessed Raman spectral image; the machine vision module is configured to analyze the high specificity of the spectral characteristics and determine the molecular vibration information of the biological sample; extracting disease diagnosis features from the Raman spectrum data according to molecular vibration information of the biological sample and the image features; identifying and classifying the disease diagnosis features according to a pre-constructed model, and determining a pathological state; and the auxiliary diagnosis module is configured to output a structured diagnosis report and a diagnosis result in combination with clinical information and the pathological state. According to the diagnosis system provided by the invention, pathological diagnosis can be quickly, comprehensively and accurately carried out.

Description

Technical Field

[0001] The present invention relates to the field of Raman spectroscopy technology, and in particular, to a diagnostic system and method. Background Art

[0002] Pathological diagnosis is a very important part of medical diagnosis. It can perform detailed typing and grading of tumors and is regarded as the "gold standard" in tumor diagnosis. Traditional pathological diagnosis procedures mainly include the processing of tissue specimens; the observation of pathological sections; the reading and diagnosis by pathologists; and the diagnosis of diseases through the morphological and structural changes of tissue cells. The existing limitations and disadvantages are as follows: Local sections cannot fully represent the entire lesion; the diagnosis process takes a long time; and it has high requirements for the professional level of technicians. Summary of the Invention

[0003] The purpose of the embodiments of the present invention is to provide a diagnostic system and method, which can solve at least one of the above problems existing in the existing case diagnosis scheme.

[0004] To solve the above technical problems, the present invention provides the following technical solutions: The embodiments of the present invention provide a diagnostic system, wherein the system includes: a data acquisition module, an image processing module, a feature acquisition module, a machine vision module, and an auxiliary diagnosis module; The data acquisition module is configured to acquire a Raman spectrum image of a target tissue; The image processing module is configured to preprocess the Raman spectrum image to obtain a preprocessed Raman spectrum image; The feature acquisition module is configured to extract spectral features and image features from the preprocessed Raman spectrum image; The machine vision module is configured to analyze the high specificity of the spectral features to determine the molecular vibration information of a biological sample; extract disease diagnosis features from the Raman spectrum data based on the molecular vibration information of the biological sample and the image features; and identify and classify the disease diagnosis features according to a pre-constructed model to determine the pathological state; The auxiliary diagnosis module is configured to combine clinical information and the pathological state to output a structured diagnostic report and a diagnostic result.

[0005] Optionally, the data acquisition module includes: An acquisition sub-module configured to acquire a tissue section of a biological sample; A placement sub-module configured to place the target tissue in a sample pool and ensure the stability of the sample and appropriate lighting conditions; A collection sub-module, configured to scan the biological sample using a Raman spectrometer to collect the molecular vibration spectrum of the biological sample and the chemical composition information of the biological sample; A generation sub-module, configured to generate a Raman spectral image of the target tissue based on the molecular vibration spectrum and the chemical composition information of the biological sample.

[0006] Optionally, the image processing module includes: A first sub-module, configured to perform baseline correction and smoothing processing on the collected Raman spectral image; A second sub-module, configured to extract the features of the smoothed Raman spectral image by principal component analysis; A third sub-module, configured to generate spatially resolved spectral information of the sample by grid measurement based on the features of the Raman spectral image; A fourth sub-module, configured to perform polynomial fitting for baseline correction; A fifth sub-module, configured to reduce the noise of the spectral information using a preset filter; A sixth sub-module, configured to adjust the intensity ratio of the spectral information using normalization to obtain a preprocessed Raman spectral image.

[0007] Optionally, the feature acquisition module is specifically configured to: Successively extract spectral features and image features in the preprocessed Raman spectral image through threshold processing and edge detection techniques; Obtain molecular vibration information by using the highly specific binding of Raman spectroscopy and chemometric methods.

[0008] Optionally, the machine vision module is specifically configured to: Use the principal component analysis method to perform dimensionality reduction processing on the spectral features; Apply the multi-curve analysis method to analyze the spectral features after dimensionality reduction processing; Adopt the partial least squares regression PLSR algorithm to perform directional analysis on the spectral features after analysis; Use the LSTM-AE model to perform feature learning and classification model training on the spectral features after analysis.

[0009] Optionally, a personalized medical advice generation algorithm is preset in the auxiliary diagnosis module; the personalized medical advice generation algorithm generates a treatment plan and health management advice for the patient according to the patient's pathological state, clinical information, and demographic characteristics.

[0010] Optionally, the diagnostic system further includes: a user interaction module configured to: Display a diagnostic parameter setting interface; receive patient information and diagnostic parameters input by the user; Visualize the diagnostic report, the diagnostic result, the treatment plan, and the health management advice.

[0011] Optionally, when the machine vision module extracts disease diagnosis features from the Raman spectroscopy data based on the molecular vibration information and the image features of the biological sample, it is specifically configured to: Combine the molecular vibration information of the biological sample with machine vision technology, and combine machine learning and deep learning algorithms to extract disease diagnosis features from the Raman spectroscopy data.

[0012] An embodiment of the present invention provides a diagnostic method, where the method includes: Obtain a Raman spectroscopy image of a target tissue; Preprocess the Raman spectroscopy image to obtain a preprocessed Raman spectroscopy image; Extract spectral features and image features from the preprocessed Raman spectroscopy image; Analyze the high specificity of the spectral features to determine the molecular vibration information of the biological sample; Extract disease diagnosis features from the Raman spectroscopy data based on the molecular vibration information and the image features of the biological sample; Identify and classify the disease diagnosis features based on a pre-constructed model to determine the pathological state; Combine clinical information and the pathological state to output a structured diagnostic report and a diagnostic result.

[0013] An embodiment of the present invention also provides an electronic device, where the electronic device includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other through the communication bus. The memory is used to store a computer program. The processor is used to implement the diagnostic method as described in any of the above embodiments when executing the program stored in the memory.

[0014] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the diagnostic method as described in any of the above embodiments.

[0015] The data acquisition module in the diagnostic system provided by the embodiments of the present invention is configured to acquire Raman spectroscopic images of a target tissue; the image processing module is configured to preprocess the Raman spectroscopic images to obtain preprocessed Raman spectroscopic images; the feature acquisition module is configured to extract spectral features and image features from the preprocessed Raman spectroscopic images; the machine vision module is configured to analyze the high specificity of the spectral features to determine the molecular vibration information of a biological sample; extract disease diagnosis features from the Raman spectroscopic data based on the molecular vibration information and image features of the biological sample; identify and classify the disease diagnosis features according to a pre-constructed model to determine the pathological state; the auxiliary diagnosis module is configured to combine clinical information and the pathological state to output a structured diagnosis report and a diagnosis result. Through the diagnostic system provided by the embodiments of the present invention, pathological diagnosis can be performed quickly, comprehensively, and accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a structural block diagram of a diagnostic system according to an embodiment of the present application; Figure 2 is a schematic diagram of the working principle of a machine vision module according to an embodiment of the present application; Figure 3 is a schematic diagram of the working principle of an image processing module according to an embodiment of the present application; Figure 4 is a flowchart of the steps of a diagnostic method according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0018] The chemotherapy effect monitoring solution provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings, specific embodiments, and their application scenarios.

[0019] As shown in the appended Figure 1 drawings, the diagnostic system of the embodiments of the present application includes the following functional modules: A data acquisition module 101, an image processing module 102, a feature acquisition module 103, a machine vision module 104, and an auxiliary diagnosis module 105; The data acquisition module 101 is configured to acquire Raman spectroscopic images of a target tissue; In an optional embodiment, the data acquisition module includes the following sub-functional modules: An acquisition sub-module configured to acquire tissue sections of a biological sample; A placement sub-module configured to place the target tissue in a sample cell and ensure the stability of the sample and appropriate lighting conditions; A collection sub-module, configured to scan a biological sample using a Raman spectrometer to collect the molecular vibration spectrum of the biological sample and the chemical composition information of the biological sample; A generation sub-module, configured to generate a Raman spectrum image of a target tissue based on the molecular vibration spectrum and the chemical composition information of the biological sample.

[0020] The data acquisition module further includes a Raman spectrometer for scanning a biological sample and collecting its molecular vibration spectrum and chemical composition information, and the Raman spectrometer cooperates with each sub-function module.

[0021] An image processing module 102, configured to preprocess the Raman spectrum image to obtain a preprocessed Raman spectrum image; In an optional embodiment, the image processing module includes the following sub-function modules: A first sub-module, configured to perform baseline correction and smoothing processing on the acquired Raman spectrum image; A second sub-module, configured to extract the features of the smoothed Raman spectrum image by principal component analysis; A third sub-module, configured to generate spectroscopically resolved spectral information of the sample in the sample space by grid measurement based on the features of the Raman spectrum image; A fourth sub-module, configured to perform polynomial fitting for baseline correction; A fifth sub-module, configured to use a preset filter to reduce the noise of the spectral information; for example, using a Savitzky-Golay filter; A sixth sub-module, configured to adjust the intensity ratio of the spectral information by normalization method to obtain a preprocessed Raman spectrum image, for example, using Min-Max normalization method.

[0022] The image processing module includes algorithms for performing operations such as baseline correction, noise reduction, normalization, derivative spectroscopy, deconvolution, etc. Each algorithm is implanted into each sub-module included in the image processing module to execute the corresponding algorithm logic.

[0023] A feature acquisition module 103, configured to extract spectral features and image features from the preprocessed Raman spectrum image; In an optional embodiment, the feature acquisition module is specifically configured to: sequentially extract spectral features and image features from the preprocessed Raman spectrum image through threshold processing and edge detection techniques; and obtain molecular vibration information by highly specific binding of Raman spectrum and chemometric methods.

[0024] The feature acquisition module includes algorithms for performing image processing techniques such as threshold processing and edge detection, as well as chemometric methods for obtaining molecular vibration information of biological samples from the preprocessed images. Each algorithm included in the feature acquisition module is used to support the corresponding functions of the feature acquisition module.

[0025] The machine vision module 104 is configured to analyze the high specificity of spectral features to determine the molecular vibration information of biological samples; extract disease diagnosis features from Raman spectral data based on the molecular vibration information and image features of biological samples; identify and classify the disease diagnosis features according to a pre-constructed model to determine the pathological state.

[0026] A feasible implementation method is that when the machine vision module extracts disease diagnosis features from Raman spectral data based on the molecular vibration information and image features of biological samples, it is specifically configured to: combine the molecular vibration information of biological samples with machine vision technology, and combine machine learning and deep learning algorithms to extract disease diagnosis features from Raman spectral data.

[0027] In an optional embodiment, the machine vision module is specifically configured to: use the principal component analysis method to perform dimensionality reduction processing on spectral features; apply the multivariate curve analysis method to analyze the spectral features after dimensionality reduction processing; use the partial least squares regression PLSR algorithm to perform directional analysis on the analyzed spectral features; use the LSTM - AE model to perform feature learning and classification model training on the analyzed spectral features.

[0028] The machine vision module includes machine learning and deep learning algorithms for feature learning and model training, such as support vector machines, convolutional neural networks, etc. It also includes feature extraction algorithms such as principal component analysis (PCA), independent component analysis (ICA), or manifold learning (UMAP) for extracting key features from the preprocessed spectral data. The machine vision module also includes an adaptive learning algorithm for dynamically adjusting model parameters according to the feature distribution of Raman spectral data to optimize the accuracy of disease classification; an ensemble learning framework for combining the prediction results of multiple classifiers to improve the generalization ability and robustness of the model to different pathological states. Moreover, the machine vision module also includes a model training and validation process for splitting the collected data into a training set, a validation set, and a test set.

[0029] The auxiliary diagnosis module 105 is configured to combine clinical information and pathological state and output a structured diagnosis report and a diagnosis result.

[0030] In an alternative embodiment, a personalized medical advice generation algorithm is preset in the auxiliary diagnosis module; the personalized medical advice generation algorithm generates a treatment plan and health management advice for the patient based on the patient's pathological status, clinical information, and demographic characteristics.

[0031] The auxiliary diagnosis module can also be configured to: use a support vector machine (SVM) to learn the features and establish a classification model; select a base model as the basis for the recursive feature elimination (RFE), initialize the RFE object and set the number of features to be retained; perform model interpretation and evaluation, including cross-validation, confusion matrix visualization, model goodness-of-fit evaluation, etc.

[0032] More preferably, the diagnostic system further includes: a user interaction module configured to: display a diagnostic parameter setting interface; receive the patient information and diagnostic parameters input by the user; visualize the diagnostic report, diagnostic results, treatment plan, and health management advice.

[0033] The data acquisition module in the diagnostic system provided by the embodiment of the present invention is configured to obtain the Raman spectral image of the target tissue; the image processing module is configured to preprocess the Raman spectral image to obtain the preprocessed Raman spectral image; the feature acquisition module is configured to extract the spectral features and image features in the preprocessed Raman spectral image; the machine vision module is configured to analyze the high specificity of the spectral features to determine the molecular vibration information of the biological sample; extract the disease diagnosis features from the Raman spectral data based on the molecular vibration information and image features of the biological sample; identify and classify the disease diagnosis features according to the pre-constructed model to determine the pathological status; the auxiliary diagnosis module is configured to combine the clinical information and pathological status and output a structured diagnostic report and diagnostic results. Through the diagnostic system provided by the embodiment of the present invention, case diagnosis can be performed quickly, comprehensively, and accurately.

[0034] Next, a specific example is used to illustrate the chemotherapy effect monitoring method provided by the embodiments of the present application.

[0035] This specific example provides a pathological diagnosis intelligent system based on the combination of Raman spectroscopy and machine vision. The system structure diagram is as Figure 1 shown. The present invention provides a pathological diagnosis intelligent system combining Raman spectroscopy and machine vision technology. The system includes the following modules: A data acquisition module configured to obtain the Raman spectral image of the target tissue; by obtaining a biological sample and performing appropriate preprocessing to adapt to Raman spectroscopy analysis, using a Raman spectrometer to scan the sample and collect its molecular vibration spectrum and the chemical composition information of the sample.

[0036] The data acquisition module consists of: obtaining tissue sections of biological samples, placing the target tissue in a sample cell, ensuring the stability of the sample and appropriate lighting conditions, and performing appropriate pre-treatment to adapt to Raman spectroscopy analysis; Select a laser as the excitation light source for Raman spectroscopy analysis in the wavelength range from ultraviolet, visible to near-infrared; Use a Raman spectrometer to scan the sample and collect its molecular vibration spectrum and chemical composition information of the sample; Use a Raman spectrometer to measure the target tissue, collect the scattered light, and record the wavelength position and intensity of the Raman scattered light to form a Raman spectrum.

[0037] The image processing module is configured to pre-process the Raman spectrum image to obtain a pre-processed Raman spectrum image; perform operations including baseline correction, noise reduction, normalization, derivative spectrum, deconvolution, etc. on the acquired Raman spectrum image to obtain a pre-processed Raman spectrum image. Figure 3 It is a schematic diagram of the workflow of the image processing module, which includes steps such as baseline correction, smoothing processing, and feature extraction.

[0038] The image processing module consists of: performing baseline correction and smoothing processing on the Raman spectrum data of the collected tissue to eliminate the influence of background noise and baseline drift; Extract the features of the Raman spectrum data by principal component analysis to identify and classify different biological tissues; Use grid measurement (10 x 10μm in size) to generate spatially resolved spectral information of the sample, thereby obtaining Raman imaging. In the Raman image, each pixel contains molecular information and consists of a complete Raman spectrum.

[0039] Analyze the Raman spectrum image and identify the characteristic peaks, which represent the wavelength position and intensity of the corresponding Raman scattered light and correspond to specific molecular bond vibrations; In particular, for tissues such as skin, obtain a target distribution map image through multivariate analysis based on Raman spectroscopy to study the distribution of transdermal targets in the skin; Perform polynomial fitting for baseline correction. By selecting data points in the non-peak region and fitting a quadratic polynomial, create a mathematical model to estimate the baseline. And subtract this polynomial model from the original spectral data to obtain the corrected spectral data. Specifically, use Python code to implement this process, where the numpy library is used for data operations and the matplotlib library is used for plotting and display; identify the baseline by calculating the first derivative of the spectral data. The first derivative of the baseline is close to zero, and the baseline is estimated by finding the points where the derivative is close to zero.

[0040] Use the Savitzky-Golay filter for noise reduction to smooth the data by fitting a polynomial to the data and calculating its value at each point. It can effectively reduce noise while preserving the shape and features of the signal.

[0041] For normalization, use the Min-Max normalization method, specifically: x* =(x−min) / (max− min ), where max is the maximum value of the sample data and min is the minimum value of the sample data. Adjust the intensity ratio of the spectrum to the same level, which helps in the comparison between data.

[0042] Based on the above operations, specifically obtain the first derivative by calculating the difference between two consecutive points, which reveals the start and end positions of the peaks and identifies the exact positions of the peaks; obtain the second derivative by calculating the derivative of the first derivative to distinguish very close peaks. Highlight the details of the spectral curve, separate overlapping peaks, and eliminate the influence of baseline drift by calculating the first, second, or higher-order derivatives.

[0043] The feature acquisition module is configured to extract spectral features and image features from the preprocessed Raman spectral image; through image processing techniques such as threshold processing and edge detection, extract spectral features and image features from the preprocessed Raman spectral image, including peak position, peak width, peak area, texture analysis, morphological features, etc., to assist in identifying and segmenting pathological regions; utilize the high specificity of Raman spectroscopy and combine chemometric methods to obtain molecular vibration information of biological samples from the preprocessed image.

[0044] The feature acquisition module consists of: determining one or more thresholds according to the characteristics of the spectrum. These thresholds can be fixed or dynamically calculated based on the dataset by analyzing the distribution of peak intensities.

[0045] Use the threshold to identify the peaks in the spectrum; Integrate the detected peaks and calculate their areas as one of the features; Measure the width of the peaks, achieved by full width at half maximum (FWHM), and conduct peak width analysis to understand the uniformity or crystallinity of the sample; Calculate the ratio between different peaks to provide information about the chemical composition of the sample; Specifically, for non-uniform images, adopt the adaptive threshold in the local threshold method to determine the threshold according to the local characteristics of the image.

[0046] Use multiple thresholds to segment the image into multiple regions, each region corresponding to different chemical components or structures; After threshold processing, morphological operations are applied. The specific operations include: 1. Selecting a structuring element; 2. Using the structuring element to perform operations such as dilation, erosion, opening, and closing on the image; 3. Executing through iterative control to achieve the desired effect; 4. After morphological operations, perform some post-processing steps to remove noise or connect broken regions by removing isolated pixels and filling holes.

[0047] Furthermore, perform texture analysis on the region after threshold segmentation to extract the following texture features: Extract the following features: Angularity: Describing the roughness of the texture; Contrast: Measuring the amplitude of gray level changes; Homogeneity: Describing the uniformity of the texture; Energy: A combination of the roughness and uniformity of the texture; Correlation: Measuring the linear relationship between the gray values of pixels.

[0048] The specific operations include: 1. Selecting the relative positions of neighboring pixels (4 nearest neighbors or 8 nearest neighbors). Determining the distance between the considered pixel pairs (1 pixel wide); 2. For each pixel in the image, find its neighboring pixels in the selected direction and distance. 3. Record the gray levels of these two pixels and increment the count at the corresponding position in the GLCM; 4. Calculate various texture features from the GLCM. Furthermore, perform further image analysis on the extracted features. The specific operations include: 1. Interpreting the texture attributes of the image based on the feature values. Different features may be related to different texture characteristics; 2. Extracting the shape features of the region after threshold segmentation, such as area, perimeter, shape complexity, etc.; 3. Identifying specific structures in the sample by identifying and analyzing the connected components in the image.

[0049] A machine vision module is configured to utilize the high specificity of Raman spectroscopy to obtain the molecular vibration information of a biological sample, combine the molecular vibration information of the biological sample with machine vision technology, and combine machine learning and deep learning algorithms to extract features from Raman spectroscopy data that are helpful for disease diagnosis and build a model to automatically identify and classify different pathological states.

[0050] The machine vision module consists of: Use PCA to perform dimensionality reduction on spectral data, extract the main components, and identify patterns and structures in the data. The specific steps are as follows: 1. For the preprocessed data, center the data, that is, perform a mean subtraction operation on the spectral intensity of each wavelength (column) to make the data have a zero mean; 2. Calculate the covariance matrix or correlation coefficient matrix of the centered data to determine the correlation between wavelengths in the data; 3. Perform eigenvalue decomposition on the covariance matrix or correlation coefficient matrix to obtain eigenvalues and eigenvectors; 4. According to the magnitude of the eigenvalues, select the most important several eigenvectors. The corresponding principal components explain the largest variance in the data; 5. Project the original data onto these principal components to obtain a new dimensionality-reduced spectral dataset and transform it into a new space; 6. Visualize the results of the first few principal components using two-dimensional or three-dimensional scatter plots for easy analysis and interpretation; 7. Validate the PCA model to check whether the sample distribution in the new space is reasonable and whether sufficient information is retained.

[0051] The processes of PCA dimensionality reduction, MCR analysis, PLSR quantitative analysis, and LSTM-AE model training used in the machine vision module are as Figure 2 shown.

[0052] Apply MCR analysis to analyze complex spectral data and extract pure component spectra and their distributions in the image. The specific operations include: 1. Use an iterative method to gradually analyze the spectra. In each iteration, the algorithm attempts to optimize the pure spectra of the components and their concentration profiles in the mixture; 2. Apply non-negativity constraints; 3. After each iteration, evaluate the quality of the current solution. This can be done by checking the difference between the reconstructed spectra and the actual data; 4. Adjust the number of iterations, regularization parameters, etc. of MCR as needed to improve the results.

[0053] By establishing a calibration model, partial least squares regression (PLSR), associate spectral data with standard samples of known concentrations to achieve quantitative analysis. The specific steps are as follows: Split the preprocessed dataset into a calibration set (for model building) and a validation set (for evaluating model performance); Use feature selection techniques, principal component analysis (PCA) to select the features that contribute the most to the model; Use the calibration set data to build a model through the PLSR algorithm. Determine the optimal PLSR model parameters, such as the number of principal components (or latent variables).

[0054] Use the validation set data to evaluate the performance of the PLSR model. The performance metrics include the coefficient of determination (R²), root mean square error (RMSE), and correlation coefficient; According to the validation results, adjust the PLSR model parameters, increase or decrease the number of principal components to optimize the model performance.

[0055] The long short-term memory-autoencoder (LSTM-AE) model is used to perform feature learning on spectral multi-dimensional data. By combining the long short-term memory (LSTM) model in the neural network model, which can effectively memorize the time series correlation characteristics, and the autoencoder (AE) model with strong self-learning ability, the feature learning effect of multi-dimensional complex time series data is effectively guaranteed.

[0056] The LSTM-AE model conducts feature learning and training of the classification model on Raman spectral data, which specifically includes the following key steps: 1. The preprocessed Raman spectral data set is divided into a training set, a validation set, and a test set.

[0057] 2. Design the LSTM-AE model: 1) Define the input layer, the input interface of the model, and specify the shape and type of the data to accept Raman spectral data; 2) Design the encoder part and use LSTM layers to extract features: Design multiple LSTM layers to gradually reduce the dimension of the data while extracting features; 3) Add dropout layers between the LSTM layers to reduce overfitting; 4) Design the bottleneck layer to further compress the feature dimension: The bottleneck layer is a fully connected layer in the autoencoder, located between the output of the encoder and the input of the decoder. Its purpose is to further compress the features, reduce the dimension of the data, and retain the most important information as much as possible; Select a number of units smaller than the output dimension of the encoder to achieve feature compression. Add the ReLU non-linear activation function to introduce non-linearity and help the model learn more complex feature representations; 5) Design the decoder part and use LSTM layers symmetric to the encoder to reconstruct the input data. This design allows the model to learn an effective representation of the data and reconstruct the data through the decoder part, thereby realizing automatic feature learning and denoising or compression of the data.

[0058] 3. Select the mean squared error (MSE) as the loss function, calculate the average of the squares of the differences between the model prediction values and the actual values, and encourage the model output to be as close as possible to the reconstruction of the original input.

[0059] 4. Select the Adam optimizer: As an optimization algorithm with an adaptive learning rate, Adam combines the advantages of the RMSProp and Momentum optimization algorithms and performs well in training deep learning models. It provides settings for some key hyperparameters, such as the learning rate, β1 (the decay rate of the first moment estimate), β2 (the decay rate of the second moment estimate), and ε (a small constant for numerical stability), and automatically adjusts the learning rate to adapt to different parameters.

[0060] 5. Train the LSTM-AE model using the training set data: Further divide the training set into smaller batches to fit the memory and the training process; use the batch data of the training set to train the model and record the loss values during the training process.

[0061] 6. Use the validation set data to adjust hyperparameters and prevent overfitting: Use cross-validation or the performance based on the validation set to adjust the hyperparameters, and apply techniques such as dropout and L2 regularization to reduce overfitting.

[0062] 7. Use the encoder part of the trained LSTM-AE model to extract features: Propagate the data forward through the encoder part to obtain the output of the bottleneck layer, and select the output of the bottleneck layer as the feature input to the classification model.

[0063] 8. Design a classification model based on the extracted features, such as SVM, random forest, or deep neural network.

[0064] 9. Adjust the hyperparameters of the LSTM-AE and classification models according to the evaluation results.

[0065] The auxiliary diagnosis module is configured to combine relevant clinical information and output a structured report and results. Build an intelligent diagnosis model to automatically identify and classify different pathological states and achieve rapid and accurate diagnosis of diseases. The model outputs a diagnosis result based on the Raman spectroscopic image of the test sample and organizes the qualitative disease classification or quantitative biochemical component analysis into a structured report.

[0066] The auxiliary diagnosis module includes: Use the support vector machine (SVM) to learn the features and establish a classification model. The specific steps are as follows: 1. Divide the dataset into a training set and a test set, and use the cross-validation method to evaluate the model performance; 2. Select a suitable SVM model according to the task type, and use the SVM classifier for classification problems; 3. Set kernel functions such as linear kernel, polynomial kernel, radial basis function (RBF) kernel, and Sigmoid kernel; 4. Use the training set data to train the SVM model. During the training process, the SVM will find the optimal hyperplane that maximizes the classification boundary; 5. Use parameter optimization techniques such as Grid Search or Random Search to find the best model parameters; 6. Use the test set or cross-validation method to evaluate the model performance; 7. Apply the trained SVM model to a new dataset for prediction.

[0067] Select a base model as the basis for RFE, initialize the RFE object, select the base model, and set the number of features to be retained; use the RFE method to recursively build the base model, removing the features with the smallest weights in each iteration until the set number of features is reached; train the base model on the selected feature subset; use appropriate evaluation metrics to evaluate the performance of the model; use techniques such as cross-validation and grid search to tune the parameters of the base model; use an independent validation set or cross-validation method to verify the performance of the model; For the above model, perform model interpretation and evaluation. The specific steps are as follows: 1. Use cross-validation to evaluate the stability and generalization ability of the model; 2. Visualize the classification performance of the model through a confusion matrix. 3. Model goodness of fit; for regression problems, use the R² score or mean squared error (MSE) to evaluate the goodness of fit of the model. 4. Check the distribution of the residuals (the differences between the actual values and the predicted values) to understand whether the model follows the linear assumption. 5. Plot learning curves to evaluate how the performance of the model changes as the number of training samples increases. 6. Plot validation curves to evaluate the impact of model parameters on performance. 7. According to the evaluation results, consider whether more data needs to be collected or the model needs to be adjusted. 8. Analyze cases where the model makes prediction errors to understand the limitations of the model. 9. Deploy the model to actual applications and monitor its performance.

[0068] Furthermore, apply the model to actual clinical samples for further testing and validation to promote the clinical translation of the model.

[0069] Next, the specific implementation executed by each functional module of the diagnostic system will be described with specific embodiments.

[0070] The implementation of the data acquisition module is as follows: Biological sample preparation: Select the target biological tissue sample and perform appropriate pre-processing, such as fixation, sectioning, etc., to adapt to Raman spectroscopy analysis.

[0071] Sample placement: Place the processed tissue sections in a dedicated sample cell and ensure the stability of the samples and appropriate lighting conditions.

[0072] Raman spectroscopy scanning: Use a Raman spectrometer to scan the samples and collect molecular vibration spectra and chemical composition information of the samples.

[0073] The implementation of the image processing module is as follows: Baseline correction: Perform baseline correction on the collected Raman spectroscopy data to eliminate the influence of background noise and baseline drift.

[0074] Smoothing processing: Use a Savitzky-Golay filter to smooth the data and reduce noise.

[0075] Feature extraction: Extract the features of the Raman spectroscopy data through principal component analysis (PCA) to provide key information for subsequent disease diagnosis.

[0076] The implementation of the feature acquisition module is as follows: Image feature extraction: Apply threshold processing and edge detection techniques to extract image features from the preprocessed Raman spectroscopy images.

[0077] Chemometric analysis: Combine chemometric methods to obtain molecular vibration information of biological samples from the images.

[0078] The implementation of the machine vision module is as follows: Dimensionality reduction processing: Use PCA to perform dimensionality reduction on the spectroscopy data, extract the main components, and simplify the data structure.

[0079] Quantitative analysis: Establish a quantitative analysis model through PLSR to associate the spectroscopy data with standard samples of known concentrations.

[0080] Deep learning model training: Use the LSTM-AE model to perform feature learning and classification model training on the spectroscopy data to improve the accuracy of diagnosis.

[0081] The implementation of the auxiliary diagnosis module is as follows: Classification model construction: Use machine learning algorithms such as SVM to establish a disease classification model based on the extracted features.

[0082] Feature selection: Adopt methods such as RFE for feature selection to optimize the model performance.

[0083] Model evaluation: Evaluate the model through methods such as cross-validation and confusion matrix visualization to ensure the stability and accuracy of the model.

[0084] The clinical sample testing process is as follows: Sample collection: Collect clinical samples and perform pathological diagnosis using the system of the present invention.

[0085] Model application: Apply the trained model to the Raman spectroscopy image of the clinical sample and output the diagnostic result.

[0086] Result analysis: Analyze the diagnostic result output by the model in combination with clinical information and provide a structured report.

[0087] The above embodiments are only examples of the present invention, and can be adjusted and optimized according to different sample characteristics and diagnostic requirements in actual applications.

[0088] Figure 4 It is a step flowchart of a diagnostic method for implementing an embodiment of the present application.

[0089] The diagnostic method provided by the embodiment of the present invention includes the following steps: Step 201: Obtain the Raman spectroscopy image of the target tissue; Step 202: Preprocess the Raman spectroscopy image to obtain the preprocessed Raman spectroscopy image; Step 203: Extract spectral features and image features from the preprocessed Raman spectroscopy image; Step 204: Analyze the high specificity of the spectral features to determine the molecular vibration information of the biological sample; Step 205: Extract disease diagnosis features from the Raman spectroscopy data based on the molecular vibration information of the biological sample and the image features; Step 206: Identify and classify the disease diagnosis features according to the pre-constructed model to determine the pathological state; Step 207: Combine clinical information and pathological state to output a structured diagnostic report and a diagnostic result.

[0090] Each step included in the diagnostic method provided in the embodiment of the present application corresponds to each functional module in the diagnostic system in the foregoing embodiment respectively. For specific implementation, refer to the relevant descriptions in the diagnostic system. To avoid repetition, it will not be elaborated here.

[0091] The embodiment of the present invention also provides an electronic device, including a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus.

[0092] The memory is used to store a computer program; The processor is used to implement the operations performed by each functional module in the above system embodiment when executing the program stored in the memory.

[0093] The communication bus mentioned in the above terminal may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0094] The communication interface is used for communication between the above terminal and other devices.

[0095] The memory may include a Random Access Memory (RAM), or may also include a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0096] The above-mentioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0097] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0098] It should be noted that in this document, the term "including", "comprising", or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article, or device that includes a series of elements includes not only those elements but also other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article, or device that includes the element.

[0099] The above are the preferred embodiments of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A diagnostic system, characterized in that: The system includes: a data acquisition module, an image processing module, a feature acquisition module, a machine vision module and an auxiliary diagnosis module; A data acquisition module configured to acquire a Raman spectrum image of a target tissue; An image processing module is configured to preprocess the Raman spectrum image to obtain a preprocessed Raman spectrum image; A feature acquisition module, configured to extract spectral features and image features from the preprocessed Raman spectral image; A machine vision module is configured to analyze the high specificity of the spectral features to determine the molecular vibration information of the biological sample; extract disease diagnostic features from the Raman spectral data based on the molecular vibration information of the biological sample and the image features; identify and classify the disease diagnostic features based on a pre-built model to determine the pathological state; The auxiliary diagnosis module is configured to combine the clinical information and the pathological state to output a structured diagnosis report and diagnosis results.

2. The diagnostic system according to claim 1, characterized in that The data acquisition module comprises: an acquisition submodule configured to acquire tissue sections of biological samples; a placement submodule configured to place the target tissue in the sample pool and ensure stability and proper lighting conditions of the sample; A collecting submodule, configured to use a Raman spectrometer to scan the biological sample, and collect the molecular vibration spectrum of the biological sample and the chemical composition information of the biological sample; The generating submodule is configured to generate a Raman spectrum image of the target tissue according to the molecular vibration spectrum and the chemical composition information of the biological sample.

3. The diagnostic system according to claim 1, characterized in that The image processing module comprises: The first submodule is configured to perform baseline correction and smoothing processing on the collected Raman spectrum image; The second submodule is configured to extract features of the smoothed Raman spectrum image by principal component analysis; A third submodule is configured to generate spatially resolved spectral information of the sample using grid measurement according to the characteristics of the Raman spectral image; a fourth submodule configured to perform polynomial fitting for baseline correction; A fifth submodule is configured to reduce noise of the spectral information using a preset filter; The sixth submodule is configured to apply a normalization method to adjust the intensity ratio of the spectral information to obtain a preprocessed Raman spectral image.

4. The diagnostic system according to claim 1, characterized in that The feature acquisition module is specifically configured as follows: Extracting spectral features and image features from the preprocessed Raman spectral image by threshold processing and edge detection technology in turn; The high specificity of Raman spectroscopy is combined with chemometric methods to obtain molecular vibration information.

5. The diagnostic system according to claim 1, characterized in that The machine vision module is specifically configured as follows: Using principal component analysis method to reduce the dimension of the spectral features; Applying multivariate curve analysis method to analyze the spectrum characteristics after dimension reduction processing; The partial least squares regression (PLSR) algorithm is used to perform directional analysis on the analyzed spectral features; The LSTM-AE model is used to perform feature learning and classification model training on the analyzed spectral features.

6. The diagnostic system according to claim 1, characterized in that The auxiliary diagnosis module is preset with a personalized medical advice generation algorithm; The personalized medical advice generation algorithm generates treatment plans and health management recommendations for the patient based on the patient's pathological condition, clinical information and demographic characteristics.

7. The diagnostic system according to claim 6, characterized in that The diagnostic system further includes: a user interaction module configured as: Display the diagnostic parameter setting interface; receive patient information and diagnostic parameters input by the user; The diagnostic report, the diagnostic result, the treatment plan, and the health management suggestion are visualized.

8. The diagnostic system according to claim 1, characterized in that When the machine vision module extracts disease diagnosis features from the Raman spectrum data based on the molecular vibration information of the biological sample and the image features, the machine vision module is specifically configured as follows: The molecular vibration information of the biological sample is combined with machine vision technology, combined with machine learning and deep learning algorithms, to extract disease diagnosis features from the Raman spectroscopy data.

9. A diagnostic method, characterized in that: The method comprises: Acquire Raman spectrum images of target tissues; Preprocessing the Raman spectrum image to obtain a preprocessed Raman spectrum image; Extracting spectral features and image features from the preprocessed Raman spectral image; Analyzing the high specificity of the spectral features to determine molecular vibration information of the biological sample; Extracting disease diagnosis features from the Raman spectrum data based on the molecular vibration information of the biological sample and the image features; Identifying and classifying the disease diagnostic features according to a pre-built model to determine the pathological state; Combining clinical information with the pathological condition, a structured diagnosis report and diagnosis results are output.

10. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the diagnostic method according to any one of claims 1 to 8 when executing the program stored in the memory.

Images