Lung cancer phenotype prediction auxiliary analysis method and system based on artificial intelligence

Abstract

The invention relates to the technical field of prediction analysis, and discloses a lung cancer phenotype prediction auxiliary analysis method and system based on artificial intelligence. The method comprises the following steps: carrying out adaptive preprocessing on a digital image of a lung cancer pathological section to obtain a standardized preprocessed image; carrying out hierarchical progressive tissue segmentation on the preprocessed image to obtain a lung cancer region labeled image; carrying out multi-dimensional feature cascade extraction on the lung cancer region labeled image to obtain a phenotypic feature set; performing hierarchical prediction on the phenotype feature set to obtain an immune phenotype prediction result; performing subtype mapping analysis on the immune phenotype prediction result to obtain a patient diagnosis report; and carrying out multi-dimensional data integration on the diagnosis report of the patient to obtain a personalized treatment scheme. The accuracy and interpretability of lung cancer phenotype prediction are improved through cascade extraction and hierarchical prediction modeling of multi-dimensional features.

Description

Technical Field

[0001] The present application relates to the field of predictive analysis, and in particular to an artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method and system. Background Art

[0002] In the diagnosis and treatment of lung cancer, accurate prediction of lung cancer phenotype is of great significance for the development of personalized treatment plans. Traditional lung cancer phenotype prediction mainly relies on pathologists through visual observation and experience judgment. This method has problems such as strong human subjectivity and time-consuming and labor-intensive. With the rapid development of artificial intelligence technology, advanced algorithms such as deep learning have shown broad application prospects in the automatic recognition and analysis of digital pathology slice images of lung cancer. Existing artificial intelligence technology has been able to realize the automatic recognition and analysis of digital pathology slice images of lung cancer, including cell segmentation, feature extraction and classification, which has improved the accuracy and efficiency of lung cancer phenotype prediction to a certain extent.

[0003] However, the existing technology still has many shortcomings: first, it is difficult to obtain high-quality annotated data. The annotation of lung cancer pathology section images requires professional pathologists, and the annotation process is cumbersome and time-consuming; second, the existing AI-based lung cancer phenotype prediction models often rely on specific data sets for training, which results in the model's generalization ability being limited when applied to different hospitals or different batches of pathology sections; third, although the AI ​​model can give prediction results, it often lacks sufficient interpretability, making it difficult for doctors and patients to understand and accept the prediction results; finally, the existing phenotypic prediction methods mainly focus on the characteristics of the tumor itself, while ignoring the individual differences and clinical manifestations of patients, resulting in the accuracy and practicality of personalized treatment recommendations to be improved. Summary of the invention

[0004] The present application provides an artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method and system, which is used to improve the accuracy and interpretability of lung cancer phenotype prediction through cascade extraction of multidimensional features and hierarchical predictive modeling, and at the same time combine the individual characteristics of patients to provide clinicians with more scientific and reasonable personalized treatment recommendations.

[0005] In a first aspect, the present application provides an artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method, which includes: generating a standardized preprocessed image by performing an adaptive preprocessing operation on a digital image of a lung cancer pathological section; receiving the standardized preprocessed image and performing a hierarchical progressive tissue segmentation process to generate a lung cancer region annotation image; reading the lung cancer region annotation image and performing a multidimensional feature cascade extraction operation to output a phenotypic feature set; performing a hierarchical prediction calculation based on the phenotypic feature set to generate an immune phenotype prediction result; reading the immune phenotype prediction result and performing a subtype mapping analysis to output a patient diagnosis report; receiving the patient diagnosis report and performing a multidimensional data integration process to generate a personalized treatment plan.

[0006] In a second aspect, the present application provides an artificial intelligence-based lung cancer phenotype prediction auxiliary analysis system, the artificial intelligence-based lung cancer phenotype prediction auxiliary analysis system comprising:

[0007] A processing module, used for generating a standardized pre-processed image by performing an adaptive pre-processing operation on the digital image of the lung cancer pathological section;

[0008] A segmentation module, used for receiving a standardized pre-processed image and performing hierarchical progressive tissue segmentation processing to generate a lung cancer region annotated image;

[0009] An extraction module is used to read the annotated image of the lung cancer region and perform multi-dimensional feature cascade extraction operations to output a phenotypic feature set;

[0010] A prediction module, used to perform hierarchical prediction calculations based on the phenotypic feature set to generate immune phenotype prediction results;

[0011] A mapping module, which is used to read the immunophenotype prediction results and perform subtype mapping analysis to output a patient diagnosis report;

[0012] The integration module is used to receive patient diagnosis reports and perform multi-dimensional data integration processing to generate personalized treatment plans.

[0013] In the technical solution provided in the present application, the digital images of lung cancer pathological sections are processed by adaptive preprocessing technology, which effectively improves the image quality and feature expression. At the same time, the hierarchical progressive tissue segmentation technology realizes the accurate labeling of lung cancer areas, significantly improving the accuracy and reliability of lesion area identification. The multidimensional feature cascade extraction technology can comprehensively capture the morphological, texture and color characteristics of lung cancer tissues, providing rich feature expressions for phenotype prediction, and the hierarchical predictive modeling technology makes full use of the extracted feature information to construct a stable and reliable prediction model, greatly improving the accuracy of the prediction results. In addition, the subtype mapping analysis technology can not only accurately determine the molecular subtype of lung cancer, but also provide a clear basis for determination, enhancing the interpretability of the prediction results. Finally, the multidimensional data integration technology combines the prediction results with clinical knowledge to generate a scientific and reasonable personalized treatment plan, providing strong support for clinical decision-making, thereby achieving high precision, high efficiency and high reliability of lung cancer phenotype prediction as a whole, significantly improving the scientificity and accuracy of lung cancer diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0015] Figure 1 This is a schematic diagram of an embodiment of the lung cancer phenotype prediction auxiliary analysis method based on artificial intelligence in the embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an embodiment of an artificial intelligence-based lung cancer phenotype prediction auxiliary analysis system in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The present application embodiment provides a lung cancer phenotype prediction auxiliary analysis method and system based on artificial intelligence. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "including" or "having" and any variation thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the lung cancer phenotype prediction auxiliary analysis method based on artificial intelligence includes:

[0019] Step S101, generating a standardized preprocessed image by performing an adaptive preprocessing operation on a digital image of a lung cancer pathology section;

[0020] Step S102, receiving the standardized pre-processed image and performing hierarchical progressive tissue segmentation processing to generate a lung cancer region labeled image;

[0021] Step S103, reading the annotated image of the lung cancer region and performing a multi-dimensional feature cascade extraction operation to output a phenotype feature set;

[0022] Step S104, performing hierarchical prediction calculation according to the phenotypic feature set to generate an immune phenotype prediction result;

[0023] Step S105, reading the immunophenotype prediction result and performing subtype mapping analysis, and outputting a patient diagnosis report;

[0024] Step S106: Receive the patient's diagnosis report and perform multi-dimensional data integration processing to generate a personalized treatment plan.

[0025] It is understandable that the execution subject of the present application can be an artificial intelligence-based lung cancer phenotype prediction auxiliary analysis system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.

[0026] Specifically, the digital images of lung cancer pathological sections were adaptively preprocessed. This process used a Gaussian filter to perform preliminary noise reduction on the image, filtered out random noise while retaining image detail information, and then enhanced the image contrast through an adaptive equalization algorithm, making the tissue structure clearer and more recognizable. Subsequently, the image was HE stained and separated, and the original image was decomposed into an H channel image and an E channel image, representing the staining information of the nucleus and cytoplasm, respectively. The two channels were color corrected by a color standardization reference template to ensure the color consistency of different batches of images. The corrected H channel and E channel images were merged through weighted fusion, and the image details were enhanced by non-local mean filtering, and finally a standardized preprocessed image was obtained. When the preprocessed image was hierarchically and progressively segmented, the region growing algorithm was first used for preliminary region division, and the growth region was determined based on the similarity of pixel gray values. The boundary point set was extracted and curve fitting was performed to obtain a preliminary description of the tissue contour. The geometric features of the contour curve were calculated, including indicators such as area, perimeter, and roundness, and the regions that obviously did not conform to the morphological characteristics of lung cancer tissue were removed by threshold screening. The regional connectivity was analyzed by morphological operations, and adjacent small regions were merged into a complete target region. The density clustering algorithm is used to further merge the candidate regions, remove isolated noise regions, and refine the region boundaries. Through the region completion and label propagation algorithms, an accurate annotated image of the lung cancer region is obtained.

[0027] In the multidimensional feature cascade extraction stage, nuclear morphology analysis is performed on the annotated images of lung cancer regions to calculate morphological features such as nuclear area, nuclear perimeter, and nuclear roundness to form nuclear morphology feature vectors. The mean, variance, skewness and other statistics of nuclear morphology features are calculated by statistical methods to obtain nuclear statistical features. At the same time, texture analysis is performed to calculate the energy, contrast, correlation and other characteristic components of the gray-level co-occurrence matrix to construct texture descriptors. The image is transformed into color space to analyze the density distribution characteristics of multiple channels. The nuclear statistical features, texture descriptors and color distribution features are combined and spliced ​​to form phenotypic combination features, which are reduced and standardized by principal component analysis to obtain the phenotypic feature set. In the hierarchical prediction modeling process, the phenotypic feature set is grouped and labeled, and the basic prediction features are obtained by normalization. The features are analyzed for correlation, and the features with significant correlation are selected to construct the key feature matrix. The most representative feature combination is selected by subspace decomposition and cross-validation. The features are mapped to high-dimensional space using kernel function transformation, and the confidence matrix is ​​obtained by classification scoring. Accurate immune phenotype prediction results are obtained by probability estimation and confidence threshold screening.

[0028] In the subtype mapping analysis stage, the immune phenotype prediction results are clustered to analyze the distribution of phenotypic characteristics. The subtype is initially determined by the hierarchical clustering method, and the contribution of the characteristics to the subtype determination is analyzed. The dominant characteristics are cross-validated to establish reliable subtype determination criteria. After confidence assessment and evidence synthesis, clinical diagnostic data are generated. The risk level is quantitatively evaluated to form a risk assessment result, and finally a standardized patient diagnosis report is generated. In the final multidimensional data integration stage, the patient diagnosis report is structured, key treatment elements are extracted and weights are calculated. Multi-layer treatment indicators are obtained through hierarchical decomposition, matched with clinical rules, and a set of candidate treatment plans is generated. The plans are evaluated and scored, the risk-benefit situation is analyzed, and the optimal treatment plan is selected. Drug interaction analysis is performed, drug safety strategies are formulated, dosing plans are calculated, and treatment time is arranged, and finally a complete personalized treatment plan is formed.

[0029] Taking the image processing of pathological sections of a lung cancer patient as an example, there are obvious uneven staining and noise interference in the original digital image. The image noise level is reduced by Gaussian filtering (σ=1.5), and the image contrast is increased by 30% by adaptive histogram equalization. After HE staining separation, the standard deviation of the H channel is reduced from 0.25 to 0.08, showing better color consistency. During the segmentation process, 15 suspicious areas were initially identified, and 3 lesion areas that met the morphological characteristics were retained after regional merging and boundary optimization. The feature extraction process obtained 128-dimensional nuclear morphological features, 64-dimensional texture features, and 32-dimensional color features. The feature dimension was reduced to 50 dimensions through principal component analysis, retaining 95% of the information. The confidence of the prediction model for the immune phenotype judgment of this case reached 0.92, and combined with clinical rules, a personalized treatment plan including surgical plan, targeted drug selection, and immunotherapy strategy was recommended. This complete processing flow demonstrates the effectiveness and accuracy of this method in practical applications.

[0030] In the embodiment of the present application, the digital image of the lung cancer pathological section is processed by adaptive preprocessing technology, which effectively improves the image quality and feature expression. At the same time, the hierarchical progressive tissue segmentation technology realizes the accurate labeling of the lung cancer area, significantly improving the accuracy and reliability of the lesion area identification, and the multidimensional feature cascade extraction technology can comprehensively capture the morphological, texture and color characteristics of lung cancer tissue, providing rich feature expressions for phenotype prediction, and the hierarchical predictive modeling technology makes full use of the extracted feature information to construct a stable and reliable prediction model, which greatly improves the accuracy of the prediction results. In addition, the subtype mapping analysis technology can not only accurately determine the molecular subtype of lung cancer, but also provide a clear basis for determination, enhancing the interpretability of the prediction results. Finally, the multidimensional data integration technology combines the prediction results with clinical knowledge to generate a scientific and reasonable personalized treatment plan, providing strong support for clinical decision-making, thereby achieving high precision, high efficiency and high reliability of lung cancer phenotype prediction as a whole, significantly improving the scientificity and accuracy of lung cancer diagnosis and treatment.

[0031] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0032] (1) performing a preliminary noise reduction operation on a digital image of a lung cancer pathology section by using a Gaussian filter to generate a preliminary noise reduction image, receiving the preliminary noise reduction image to perform an adaptive equalization enhancement process, and outputting an enhanced image;

[0033] (2) reading the enhanced image and executing the HE staining separation algorithm to generate an H channel image and an E channel image, receiving the H channel image and the E channel image to execute color standardization reference template correction, and outputting a standardized H channel image and a standardized E channel image;

[0034] (3) receiving the standardized H channel image and the standardized E channel image, performing a weighted fusion operation to generate a merged image, and reading the merged image to process it through a non-local mean filtering to output a detail enhanced image;

[0035] (4) reading the detail enhanced image and performing gradient threshold segmentation operation to separate the foreground and background, generating a foreground area image, receiving the foreground area image and performing morphological edge smoothing processing to output a smoothed image;

[0036] (5) Read the smoothed image to perform contrast stretching adjustment operation to generate a contrast adjusted image, receive the contrast adjusted image to perform brightness normalization processing, and output a standardized pre-processed image.

[0037] Specifically, a Gaussian filter was used for preliminary noise reduction. The Gaussian filter is a linear smoothing filter that smoothes the image by weighted averaging, where the closer the pixel is to the center pixel, the greater the weight, and vice versa. The core idea of ​​Gaussian filtering is to use the Gaussian function as a weight coefficient to perform weighted averaging on the pixel values ​​of each pixel in the image and its neighborhood. In the specific implementation, for a 1024×1024 pixel pathological image, a Gaussian filter kernel size of 5×5 was selected, and the standard deviation was set to 1.5, which effectively suppressed random noise while retaining the main structural information of the image. The noise level in the image after Gaussian filtering was significantly reduced, and the signal-to-noise ratio was increased from the original 15dB to 25dB, laying the foundation for subsequent processing.

[0038] After the initial denoising, the image then enters the adaptive equalization stage, and the mapping relationship of pixel values ​​is dynamically adjusted by calculating the histogram distribution of the local area. The adaptive equalization process divides the image into several 8×8 sub-blocks, calculates the grayscale histogram for each sub-block and performs local equalization. The bilinear interpolation method is used for smooth transition between adjacent sub-blocks to avoid the generation of block effects. This local adaptive processing method enhances the local contrast of the image while maintaining the global brightness balance. Practice shows that after adaptive equalization, the local contrast of the image is improved by an average of 40%, which effectively enhances the distinction between the cell nucleus and the cytoplasm. After obtaining the enhanced image, the HE staining separation algorithm is used to decompose the image into the H channel (hematoxylin staining) and the E channel (eosin staining). HE staining separation is based on the color anti-fading principle of the optical density space. The image in the RGB color space is converted into the optical density value using the pre-calibrated dye absorption coefficient matrix, and then the relative concentration distribution of each dye is obtained by the least squares method. In this way, the information of the cell nucleus (mainly stained by hematoxylin) and the cytoplasm (mainly stained by eosin) is separated into different channels. The separated H channel and E channel images are color corrected using a color standardization reference template, which contains the color distribution characteristics of the standard stained sample. By histogram matching, the color distribution of each channel is adjusted to be consistent with the reference template, thereby eliminating color variations caused by different staining conditions.

[0039] Next, the standardized H channel and E channel images are weighted fused. The fusion weight is determined according to the contribution of the two channels to the lesion characteristics. The H channel weight is 0.6 and the E channel weight is 0.4. The fused image retains the morphological characteristics of the nucleus and the texture information of the cytoplasm. The fused image is enhanced by non-local mean filtering. This filtering method not only considers the spatial distance of pixels, but also the similarity between pixel blocks. It can well protect the edge and texture details of the image while removing noise. In the specific implementation, a 7×7 search window and a 5×5 similarity window are selected, and the similarity parameter is set to 0.8 times the image standard deviation. The image with detail enhancement is separated from the foreground and background by gradient threshold segmentation. First, the gradient amplitude of the image is calculated, and the Sobel operator is used to calculate the horizontal and vertical gradients respectively, and then the gradient amplitude map is synthesized. According to the statistical distribution of the gradient amplitude, the value at the 75% quantile is selected as the threshold, and the area with a gradient amplitude greater than the threshold is marked as the foreground. The foreground area obtained by segmentation is smoothed by morphological operations. First, an opening operation is performed to remove small protrusions and noise points, and then a closing operation is performed to fill small holes. The operation is performed using a 3×3 structural element.

[0040] Finally, the smoothed image is contrast stretched to linearly map the grayscale range of the image to the standard interval [0, 255] to improve the visual effect of the image. Through brightness normalization, the average brightness of the image is adjusted to the standard value of 128 and the standard deviation is adjusted to 45 to ensure that different images have similar brightness distribution characteristics.

[0041] For example, the original digital image resolution is 2048×2048 pixels and the bit depth is 24-bit color. After 5×5 Gaussian filtering (σ=1.5), the peak signal-to-noise ratio of the image is increased from the original 18.3dB to 27.8dB. Adaptive equalization processing increases the local contrast by 35%, making the cell nucleus boundary clearer. The color standard deviation of the H channel and E channel images separated by HE staining is reduced from 0.23 and 0.19 to 0.08 and 0.06, respectively, reflecting good color consistency. After weighted fusion and non-local mean filtering, the structural similarity index (SSIM) of the image reaches 0.92, while maintaining important pathological features. The threshold used by gradient threshold segmentation is the 78% quantile of the gradient amplitude (value 45), and the region of interest is successfully separated. The final standardized preprocessed image has a uniform brightness distribution (mean value 128±2, standard deviation 45±1) and good contrast.

[0042] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0043] (1) Importing the preprocessed image and executing the region growing algorithm to perform preliminary region division, generate a preliminary region image, and load the preliminary region image into the boundary point extraction process to output a boundary point set;

[0044] (2) Loading the boundary point set to perform curve fitting operation, generate a contour curve, and perform geometric feature calculation on the contour curve to output the regional geometric features;

[0045] (3) Obtaining the regional geometric features and performing threshold screening processing to generate the first screening area, and importing the first screening area into the morphological operation to perform connectivity analysis and output the connected area;

[0046] (4) extracting connected regions and performing density clustering operations to merge regions, generate candidate regions, and refine the boundaries of the candidate regions to output refined boundaries;

[0047] (5) The refined boundary is loaded to perform region completion operation to generate a complete region, and the complete region is imported into the label propagation process to output a labeled image of the lung cancer region.

[0048] Specifically, the image is preliminarily divided into regions using a region growing algorithm. The region growing algorithm starts with a selected seed point and gradually expands the region based on the principle of pixel value similarity. The selection of seed points is based on the grayscale value distribution of the image, and local maximum points are selected as seed points. For each seed point, the grayscale difference between its 8 neighborhood pixels and the seed point is calculated. When the difference is less than a preset threshold (usually set to 1.5 times the local standard deviation), the neighborhood pixel is added to the growing region. This process is iterated until no new pixels meet the growth conditions. When the boundary points of the preliminary regional image are extracted, the eight-connected tracking method is used to search point by point along the edge of the region, record the coordinate information of the boundary points, and form an ordered set of boundary points. When the obtained boundary point set is subjected to curve fitting, the cubic spline interpolation method is used to fit the boundary curve through a piecewise polynomial function. The geometric feature calculation of the boundary curve includes the following indicators:

[0049] Area perimeter L:

[0050]

[0051] Where (x i ,y i ) represents the coordinates of the i-th boundary point, and n is the total number of boundary points.

[0052] Area shape complexity C:

[0053]

[0054] Where A is the area of ​​the region and L is the perimeter of the region.

[0055] Regional irregularity I:

[0056]

[0057] where r i represents the distance from the i-th boundary point to the centroid of the region, is the average distance.

[0058] The regional geometric features were threshold screened, and the shape complexity threshold range was set to [0.6, 0.9], and the irregularity threshold range was set to [0.2, 0.4] to screen out the regions that met the morphological characteristics of lung cancer tissue. The connectivity analysis of the first screened region was performed by morphological operation, and the combination of open and closed operations was used to eliminate small noise areas and fill holes. The open operation used a circular structural element with a radius of 3 pixels, and the closed operation used a circular structural element with a radius of 5 pixels to obtain a set of connected regions. The connected regions were then merged using a density clustering algorithm. Density clustering is based on the spatial distribution characteristics of the region, and the Euclidean distance and regional overlap between regions are calculated. When the distance between two regions is less than the preset threshold (set to 0.2 times the average diameter of the region) and the overlap is greater than 30%, the two regions are merged. The boundaries of the obtained candidate regions were refined, and the Snake active contour model was used to optimize the boundary position through the energy minimization criterion to obtain accurate regional boundaries.

[0059] When completing the region of fine boundaries, morphological reconstruction technology is used to fill the holes and cracks in the boundary. First, a binary mask is constructed, and then the interior of the region is gradually filled through geodesic dilation operations until a stable state is reached. Finally, the complete region is annotated by the label propagation algorithm, and the region is divided into two categories: diseased tissue and normal tissue. The label propagation process is based on the gray value distribution and spatial connection relationship of the region, and the label is gradually expanded until the entire image area is covered.

[0060] For example, the 2048×2048 pixel preprocessed image was processed. The region growing algorithm initially selected 127 seed points (based on local maximum detection), and the growth threshold was set to 12 (obtained by multiplying the local region standard deviation 8 by 1.5). The boundary extraction obtained about 8500 boundary points, and a smooth contour curve was obtained after cubic spline fitting. The geometric feature calculation results showed that the area of ​​the largest suspected lesion area was 24635 square pixels, the perimeter was 682 pixels, the shape complexity was 0.73, and the irregularity was 0.31, which met the set threshold range. After morphological operation, 23 connected regions were obtained, and density clustering merged them into 8 candidate regions. The boundary refinement process optimized the number of boundary points to an average of 2800 points per region. The region completion filled the holes with an average diameter of less than 20 pixels. Finally, the label propagation algorithm annotated these areas as lung cancer lesions, and the total area accounted for 15.7% of the original image. The consistency of the results obtained with manual annotation reached 91.3%.

[0061] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0062] (1) Loading the annotated image of the lung cancer region to perform nuclear morphology analysis operations, generate nuclear morphology feature vectors, extract the nuclear morphology feature vectors to calculate statistical parameters, and output nuclear statistical features;

[0063] (2) Import the annotated image of the lung cancer region for texture analysis and processing, establish a gray-level co-occurrence matrix, load the gray-level co-occurrence matrix to run feature component calculation, and construct a texture descriptor;

[0064] (3) Obtaining the annotated image of the lung cancer region, performing color space transformation, outputting a multi-channel image, and sending the multi-channel image to a density distribution calculation program to generate color distribution features;

[0065] (4) integrating the kernel statistical features and texture descriptors to perform feature combination operations, constructing a primary feature group, and sending the primary feature group and the color distribution features to a feature concatenation program to form a phenotypic combination feature;

[0066] (5) Import the phenotypic combination features and start principal component analysis to perform feature dimensionality reduction, output the reduced dimensionality features, and perform standardization on the reduced dimensionality features to generate a phenotypic feature set.

[0067] Specifically, nuclear morphology analysis is first performed, and single cell nucleus regions are identified and extracted through the cell nucleus segmentation algorithm, and the basic morphological features of each cell nucleus, such as area, circumference, roundness, and eccentricity, are calculated. The cell nucleus segmentation uses an adaptive threshold method combined with a watershed algorithm to separate the adherent cell nuclei. The morphological features are calculated for each cell nucleus region, and a nuclear morphology feature vector is constructed. Subsequently, statistics are calculated for the nuclear morphology feature vector, including statistical indicators such as mean, variance, skewness, and kurtosis, to form nuclear statistical features. When performing texture analysis on the annotated image of the lung cancer region, a gray-level co-occurrence matrix (GLCM) is first constructed. The gray-level co-occurrence matrix describes the spatial correlation of the grayscale values ​​of pixels in the image, and is constructed by counting the frequency of occurrence of grayscale values ​​of pixel pairs with specific spatial relationships. During the construction process, the distance of the pixel pair is set to 1, and the direction angles include four directions: 0°, 45°, 90°, and 135°. The characteristic components of the gray-level co-occurrence matrix are calculated to extract texture descriptors. Texture feature component T f The calculation formula is:

[0068]

[0069] Among them, P(m,n) represents the gray-level co-occurrence matrix element, Q(m,n) represents the normalized co-occurrence matrix, H(m,n) represents the local entropy value, μ q and μ h denote the means of Q and H respectively, σ h represents the standard deviation of H, M and N represent the number of rows and columns of the matrix respectively.

[0070] Then, the color space transformation is performed on the annotated image of the lung cancer area, and the RGB color space is converted into the HSV (hue, saturation, brightness) color space and Lab color space. The HSV space can better express human perception of color, while the Lab space has an advantage in color difference measurement. The color histogram and color moment, including the first-order moment (mean), second-order moment (variance) and third-order moment (skewness), are calculated for the converted multi-channel image to obtain the color distribution characteristics. The kernel statistical features and texture descriptors are combined, and the two groups of features are spliced ​​into a first-level feature group in a predefined order by using the feature cascade method. The importance weight of the features is considered when combining the features, and different weight coefficients are assigned to different types of features. The first-level feature group is then spliced ​​with the color distribution features to construct a complete phenotypic combination feature vector.

[0071] Finally, the phenotypic combination features are subjected to feature dimensionality reduction through principal component analysis (PCA). Principal component analysis first calculates the covariance matrix of the features, then solves the eigenvalues ​​and eigenvectors, and selects the first K principal components with the largest contribution rate as the features after dimensionality reduction. The dimensionality reduction features are standardized using the Z-score standardization method, so that the mean of the eigenvalues ​​is 0 and the standard deviation is 1, and the final phenotypic feature set is obtained.

[0072] For example, a 2048×2048 pixel region annotation image was processed, 523 cell nucleus regions were identified through nuclear morphology analysis, and the morphological characteristics of each cell nucleus were calculated. The average area was 785 square pixels, the standard deviation of the perimeter was 23.5 pixels, and the mean roundness was 0.82. The size of the constructed gray-level co-occurrence matrix was 256×256, and 20 texture feature components such as energy, contrast, and correlation were calculated. After color space transformation, 6 color channels (H, S, V, L, a, b) were obtained, and 5 statistical features were calculated for each channel, for a total of 30 color features. After feature combination, a 128-dimensional phenotypic combination feature was obtained, which was reduced to 32 dimensions through principal component analysis, retaining 95.7% of the information. The final phenotypic feature set showed good discrimination performance in the subsequent classification verification, and the classification accuracy of lung cancer tissues with different degrees of differentiation reached 89.5%.

[0073] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0074] (1) Loading the phenotypic feature set to perform grouping labeling operations, constructing grouping feature data, and sending the grouping feature data to a normalization processing program to output basic prediction features;

[0075] (2) Obtaining basic prediction features to perform correlation analysis and calculation, generate a feature correlation matrix, and initiate threshold filtering on the feature correlation matrix to construct a key feature matrix;

[0076] (3) Import the key feature matrix and run the subspace decomposition program to form a feature subspace, and then pass the feature subspace into the cross-validation system to establish a validation feature group;

[0077] (4) Load the verification feature group and perform kernel function transformation to realize feature mapping, output the mapping features, and perform classification score calculation on the mapping features to generate a confidence matrix;

[0078] (5) Extract the confidence matrix to perform probability estimation operations, output the phenotypic probability value, and import the phenotypic probability value into the confidence threshold screening program to generate the immune phenotype prediction results.

[0079] Specifically, the grouping and labeling process is first performed, and the phenotypic feature set is divided into three parts: morphological feature group, texture feature group and color feature group according to the type of feature, to generate structured grouping feature data. When the grouping feature data is normalized, the Min-Max normalization method is used to map each feature value to the [0, 1] interval, eliminate the dimension difference, and obtain the basic prediction feature.

[0080] Perform correlation analysis on basic prediction features, calculate the Pearson correlation coefficient between features, and construct a feature correlation matrix. Set a correlation threshold (usually 0.85) for the feature correlation matrix, remove highly correlated redundant features, retain features with high contribution to prediction, and obtain a key feature matrix.

[0081] When performing subspace decomposition on the key feature matrix, the following formula is used:

[0082]

[0083] Among them, S k represents the kth subspace, α j is the feature weight coefficient, V j is the eigenvector, β j is the attenuation factor, W j is the window size, D ij is the feature discrimination, J is the number of features, and I is the number of samples.

[0084] A 5-fold cross validation was performed on the feature subspace, and the data set was divided into a training set and a validation set to verify the stability and effectiveness of the subspace features and obtain a validation feature group. The validation feature group was mapped using a radial basis kernel function to map the features to a high-dimensional space to enhance the linear separability of the features. The mapped features were classified and scored using a multi-classifier ensemble method, including three base classifiers: support vector machine, random forest, and gradient boosting tree. The confidence matrix was obtained by weighted voting.

[0085] When estimating the probability of the confidence matrix, the following formula is used:

[0086]

[0087] Among them, P(C m |X) indicates that sample X belongs to category C m The probability of n is the classifier weight, E n is the classifier output, μ n is the category mean, σ nis the category standard deviation, Q is the number of classifiers, and P is the number of categories. A confidence threshold (usually 0.75) is set for the phenotype probability value to screen out the prediction results with high confidence and obtain the immune phenotype prediction results.

[0088] For example, a 32-dimensional phenotypic feature set is processed, and after grouping and marking, 12-dimensional morphological features, 8-dimensional texture features, and 12-dimensional color features are obtained. Min-Max normalization maps the feature values ​​to the interval [0, 1]. Correlation analysis found that the correlation coefficients of 7 pairs of features exceeded 0.85. After removing these highly correlated features, 25-dimensional key features were obtained. Subspace decomposition obtained 3 main feature subspaces, and cross-validation showed that the average accuracy reached 87.3%. The dimension after feature mapping was expanded to 128 dimensions, and the weights of the three base classifiers were set to 0.4, 0.35, and 0.25, respectively. The final prediction results include the degree of tumor differentiation (high differentiation probability 0.82, medium differentiation probability 0.15, low differentiation probability 0.03) and the degree of immune cell infiltration (high density probability 0.78, medium density probability 0.18, low density probability 0.04).

[0089] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0090] (1) Importing the immune phenotype prediction results to start feature clustering analysis, constructing immune feature groups, and then importing the immune feature groups into the distribution statistics program to generate phenotypic distribution data;

[0091] (2) Obtain phenotypic distribution data and perform hierarchical clustering operations to make preliminary subtype judgments, form preliminary subtype judgment results, and perform feature contribution analysis on the preliminary subtype judgment results to establish a feature importance sequence;

[0092] (3) Loading the feature importance sequence to perform sorting and screening, outputting the dominant feature set, and sending the dominant feature set to the cross-validation system to construct the subtype judgment standard;

[0093] (4) extracting the subtype determination criteria and calculating the confidence assessment parameters to generate a subtype confidence matrix, and running the evidence synthesis algorithm on the subtype confidence matrix to form clinical diagnosis data;

[0094] (5) Collect clinical diagnostic data to perform risk classification operations, establish graded risk indicators, and transmit the graded risk indicators to the quantitative assessment system to generate risk assessment results;

[0095] (6) Integrate the risk assessment results to perform diagnostic recommendation generation processing, output the diagnostic recommendation content, and apply a report formatter to the diagnostic recommendation content to construct a patient diagnostic report.

[0096] Specifically, feature clustering analysis was first performed, and the K-means clustering algorithm was used to group the immune phenotype features. The clustering process was based on the Euclidean distance measurement between features, and the cluster center was iteratively optimized to form immune feature groups with similar feature patterns. Distribution statistics were performed on the immune feature groups, and the mean, standard deviation, distribution skewness and other statistics within each feature group were calculated to form phenotypic distribution data. When the phenotypic distribution data was used for preliminary subtype judgment by hierarchical clustering, a bottom-up agglomerative hierarchical clustering method was used. Initially, each sample was regarded as an independent category, and then the closest categories were gradually merged until the preset number of categories was reached. The distance between categories was calculated using the Ward minimum variance method, and the category that could minimize the increase in intra-class variance was selected for merging. The results of hierarchical clustering formed a dendrogram structure, and the preliminary subtype classification was determined based on the branch structure of the dendrogram to obtain the preliminary subtype results. Feature contribution analysis was performed on the preliminary subtype results, and the contribution of each feature to subtype differentiation was calculated. The feature importance was measured by combining the Gini index and information gain to generate a feature importance sequence.

[0097] When sorting and screening the feature importance sequence, the importance threshold is set to the 75% quantile, and the features with importance greater than the threshold are selected to form the dominant feature set. The dominant feature set is verified by the five-fold cross-validation method. The data set is randomly divided into five subsets, and four subsets are used as training sets and one subset is used as a validation set in turn to evaluate the stability and prediction effect of the features and establish the subtype determination criteria. The confidence assessment of the subtype determination criteria is performed, and the reliability of each subtype determination result is calculated based on statistical methods to construct a subtype confidence matrix. The confidence assessment takes into account multiple aspects such as the stability of the features, the representativeness of the samples, and the consistency of the determination results. The subtype confidence matrix is ​​synthesized for evidence, and the Dempster-Shafer evidence theory is used to fuse the determination results of multiple evidence sources to generate comprehensive clinical diagnostic data.

[0098] The clinical diagnosis data is divided into risk levels, and the risks are divided into high, medium and low levels according to indicators such as tumor differentiation, invasiveness, and immune response intensity to form graded risk indicators. The graded risk indicators are quantitatively evaluated, and the weights and interactions of various indicators are comprehensively considered to calculate the comprehensive risk score to obtain the risk assessment results. Diagnostic recommendations are generated for the risk assessment results, and targeted treatment recommendations and prognosis predictions are generated based on preset clinical diagnosis and treatment rules to form diagnostic recommendation content. The diagnostic recommendation content is formatted for report, and it is typeset and organized according to the standardized medical report format to generate a standardized patient diagnosis report.

[0099] For example, the prediction results of the immune phenotype of a patient were processed. K-means clustering (K=5) divided the 32 immune phenotype features into 5 feature groups, and each feature group reflected different biological characteristics. The phenotypic distribution data showed that the first feature group (cell proliferation related) had a mean of 0.72 and a standard deviation of 0.15; the second feature group (immune cell infiltration related) had a mean of 0.45 and a standard deviation of 0.23; the third feature group (angiogenesis related) had a mean of 0.38 and a standard deviation of 0.18; the fourth feature group (cell apoptosis related) had a mean of 0.63 and a standard deviation of 0.12; the fifth feature group (matrix remodeling related) had a mean of 0.51 and a standard deviation of 0.20. Hierarchical clustering analysis classified the samples into three main subtypes. Feature contribution analysis showed that immune cell infiltration density, tumor cell proliferation index, and angiogenic factor expression level were the three main features for distinguishing subtypes, with importance of 0.82, 0.75, and 0.68, respectively. After cross-validation, the average accuracy of the subtype determination criteria reached 85.6%. The confidence level of the case was 0.87, indicating that it belonged to the immunocompetent subtype, and the clinical risk assessment result was moderate risk (risk score 68 points, full score 100 points). The final diagnostic report included detailed immunophenotypic feature analysis, molecular typing determination, prognostic risk assessment, and personalized treatment recommendations.

[0100] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0101] (1) Loading patient diagnostic reports to perform data classification and sorting, constructing classified diagnostic data, and sending the classified diagnostic data to a structured processing program to output structured diagnostic information;

[0102] (2) Collect structured diagnostic information to initiate feature correlation analysis, form key treatment factors, and apply weight calculation operations to the key treatment factors to generate a treatment factor weight matrix;

[0103] (3) Importing the treatment factor weight matrix to implement hierarchical decomposition processing, establish multi-layer treatment indicators, and then transfer the multi-layer treatment indicators to the clinical rule matching system to construct a set of candidate treatment plans;

[0104] (4) Obtaining a set of candidate treatment options for option evaluation and calculation, outputting an option scoring matrix, and running a risk-benefit analysis program on the option scoring matrix to form a feasibility evaluation report;

[0105] (5) Extract the feasibility assessment report to perform the optimal plan calculation, generate a recommended treatment list, and send the recommended treatment list to the drug interaction analysis system to establish a medication safety strategy;

[0106] (6) Integrate medication safety strategies to implement dose adjustment calculations, construct medication regimens, and initiate a time scheduling program for medication regimens to output treatment schedules;

[0107] (7) Load the treatment schedule for clinical pathway planning and form a treatment execution plan, which is then transferred to the follow-up monitoring design system to generate a personalized treatment plan.

[0108] Specifically, the diagnostic report is first classified and sorted, and the report content is divided into multiple categories such as pathological characteristics, immune phenotypes, gene mutations, clinical symptoms, etc. The classified diagnostic data is obtained through text parsing and information extraction. The classified diagnostic data is structured, and standardized data templates are used to convert unstructured text information into a standardized data format to form structured diagnostic information containing key fields and values. Feature correlation analysis is performed on the structured diagnostic information, and the degree of correlation between different diagnostic features is evaluated by correlation coefficient calculation. The key features that have an important impact on treatment decisions are screened out to form key treatment factors. The key treatment factors are weighted based on their impact on the treatment effect. The hierarchical analysis method is used to determine the relative importance of each factor, and a treatment factor weight matrix is ​​constructed.

[0109] The treatment factor weight matrix is ​​hierarchically decomposed, and the treatment factors are organized according to different dimensions and levels, including tumor characteristic layer, immune status layer, systemic condition layer, etc., to obtain multi-layer treatment indicators. The multi-layer treatment indicators are matched with the preset clinical treatment rules, and based on the treatment plan selection criteria in the rule base, a set of candidate treatment plans that meet the patient's specific conditions is screened out. A comprehensive evaluation of the candidate treatment plan set is conducted, and each plan is scored from multiple dimensions such as treatment effect, risk control, and economic cost to form a plan scoring matrix. A risk-benefit analysis is conducted on the plan scoring matrix, and the balance between treatment effect and adverse reactions is comprehensively considered to evaluate the clinical feasibility of the plan and generate a detailed feasibility evaluation report.

[0110] Optimize the feasibility assessment report, select the best treatment plan based on the comprehensive score results, and generate a recommended treatment list. Perform interaction analysis on the drug combinations in the recommended treatment list, evaluate the synergy and antagonism between drugs, formulate a drug safety strategy, and ensure the safety of the treatment plan.

[0111] Calculate the dosage adjustment for medication safety strategy and calculate the most appropriate dosage based on the patient's individual characteristics such as weight, age, liver and kidney function. Schedule the medication regimen and develop a detailed medication schedule, including medication frequency, interval time and treatment schedule. Plan the clinical pathway for the treatment schedule and design a complete treatment process, including treatment preparation, execution steps, efficacy evaluation and other links. Design a follow-up monitoring plan for the treatment execution plan, determine the monitoring indicators, follow-up frequency and warning thresholds, and ultimately form a complete personalized treatment plan.

[0112] For example, the patient's diagnosis report showed that the EGFR gene exon 20 insertion mutation was positive, the PD-L1 expression level was 60%, the immune cell infiltration was moderate, and the clinical stage was IIIA. After data classification and sorting, 28 structured diagnostic indicators were obtained, and the characteristic correlation analysis screened out 12 key treatment factors. The weight distribution calculated by the hierarchical analysis method was: gene mutation status 0.3, immune status 0.25, staging 0.2, general condition 0.15, and complications 0.1. After hierarchical decomposition, a treatment indicator system was formed, which included 3 main levels and 15 specific indicators. Clinical rule matching generated 5 candidate treatment plans. The scheme scoring results showed that the comprehensive score of scheme 1 (targeted therapy + immunotherapy) was 85 points, scheme 2 (chemotherapy + immunotherapy) was 78 points, and scheme 3 (targeted therapy alone) was 72 points. The scores of the remaining schemes were all below 70 points. The risk-benefit analysis recommended the use of scheme 1, and the drug interaction analysis showed no serious interaction risk. In the dosing regimen design, the targeted drug dose is 150 mg / day, and the immunotherapy drug dose is 200 mg / time, which is administered once every 3 weeks. The treatment schedule is arranged as follows: follow-up once a week in the first month, and then every 3 weeks, and set warning thresholds for 8 monitoring indicators.

[0113] The above describes the lung cancer phenotype prediction auxiliary analysis method based on artificial intelligence in the embodiment of the present application. The following describes the lung cancer phenotype prediction auxiliary analysis system based on artificial intelligence in the embodiment of the present application. Figure 2 In the embodiments of the present application, an embodiment of the lung cancer phenotype prediction auxiliary analysis system based on artificial intelligence includes:

[0114] A processing module, used for generating a standardized pre-processed image by performing an adaptive pre-processing operation on the digital image of the lung cancer pathological section;

[0115] A segmentation module, used for receiving a standardized pre-processed image and performing hierarchical progressive tissue segmentation processing to generate a lung cancer region annotated image;

[0116] An extraction module is used to read the annotated image of the lung cancer region and perform multi-dimensional feature cascade extraction operations to output a phenotypic feature set;

[0117] A prediction module, used to perform hierarchical prediction calculations based on the phenotypic feature set to generate immune phenotype prediction results;

[0118] A mapping module, which is used to read the immunophenotype prediction results and perform subtype mapping analysis to output a patient diagnosis report;

[0119] The integration module is used to receive patient diagnosis reports and perform multi-dimensional data integration processing to generate personalized treatment plans.

[0120] Through the coordinated cooperation of the above components, the digital images of lung cancer pathological sections are processed by adaptive preprocessing technology, which effectively improves the image quality and feature expression. At the same time, the hierarchical progressive tissue segmentation technology realizes the accurate labeling of lung cancer areas, significantly improving the accuracy and reliability of lesion area identification. The multidimensional feature cascade extraction technology can comprehensively capture the morphological, texture and color characteristics of lung cancer tissues, providing rich feature expressions for phenotype prediction, while the hierarchical predictive modeling technology makes full use of the extracted feature information to build a stable and reliable prediction model, greatly improving the accuracy of the prediction results. In addition, the subtype mapping analysis technology can not only accurately determine the molecular subtype of lung cancer, but also provide a clear basis for determination, enhancing the interpretability of the prediction results. Finally, the multidimensional data integration technology combines the prediction results with clinical knowledge to generate a scientific and reasonable personalized treatment plan, providing strong support for clinical decision-making, thereby achieving high precision, high efficiency and high reliability of lung cancer phenotype prediction as a whole, significantly improving the scientificity and accuracy of lung cancer diagnosis and treatment.

[0121] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A lung cancer phenotype prediction auxiliary analysis method based on artificial intelligence, characterized in that: The artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method includes: By performing adaptive preprocessing operations on digital images of lung cancer pathological sections, a standardized preprocessed image is generated; Receiving a standardized pre-processed image and performing hierarchical progressive tissue segmentation processing to generate a lung cancer region labeled image; Read the annotated image of the lung cancer region and perform multi-dimensional feature cascade extraction operations to output a phenotypic feature set; Performing hierarchical prediction calculations based on the phenotypic feature set to generate immune phenotype prediction results; Read the immunophenotype prediction results and perform subtype mapping analysis to output the patient diagnosis report; Receive patient diagnostic reports and perform multi-dimensional data integration processing to generate personalized treatment plans.

2. The artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method according to claim 1, characterized in that: The method of generating a standardized preprocessed image by performing an adaptive preprocessing operation on the digital image of the lung cancer pathology section comprises: Processing the digital image of the lung cancer pathology section through a Gaussian filter to perform a preliminary noise reduction operation to generate a preliminary noise reduction image, receiving the preliminary noise reduction image to perform adaptive equalization enhancement processing, and outputting an enhanced image; Read the enhanced image and perform HE staining separation algorithm operation to generate an H channel image and an E channel image, receive the H channel image and the E channel image to perform color standardization reference template correction, and output a standardized H channel image and a standardized E channel image; Receive the standardized H channel image and the standardized E channel image to perform weighted fusion operation to generate a merged image, read the merged image and process it through non-local mean filtering to output a detail enhanced image; Read the detail enhanced image and perform gradient threshold segmentation operation to separate the foreground and background, generate a foreground area image, and receive the foreground area image to perform morphological edge smoothing processing and output a smoothed image; The smoothed image is read to perform a contrast stretching adjustment operation to generate a contrast adjusted image, and the contrast adjusted image is received to perform a brightness normalization operation to output a standardized pre-processed image.

3. The artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method according to claim 1, characterized in that: The receiving of the standardized pre-processed image and performing hierarchical progressive tissue segmentation processing to generate a lung cancer region labeled image includes: Import the pre-processed image and execute the region growing algorithm to perform preliminary region division, generate a preliminary region image, load the preliminary region image into the boundary point extraction process, and output the boundary point set; Load the boundary point set to perform curve fitting operation, generate contour curve, and perform geometric feature calculation on the contour curve to output regional geometric features; Obtaining the regional geometric features to perform threshold screening processing, generate the first screening area, and import the first screening area into the morphological operation to perform connectivity analysis, and output the connected area; Extract connected regions and perform density clustering operations to merge regions, generate candidate regions, refine the boundaries of candidate regions, and output refined boundaries; The refined boundary is loaded to perform region completion operation to generate a complete region, and the complete region is imported into the label propagation process to output the annotated image of the lung cancer region.

4. The artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method according to claim 1, characterized in that: The method of reading the annotated image of the lung cancer region and performing a multi-dimensional feature cascade extraction operation to output a phenotype feature set includes: Load the annotated image of the lung cancer region to perform nuclear morphology analysis, generate nuclear morphology feature vectors, extract the nuclear morphology feature vectors to calculate statistical parameters, and output nuclear statistical features; Import annotated images of lung cancer regions for texture analysis and processing, establish a gray-level co-occurrence matrix, load the gray-level co-occurrence matrix to run feature component calculations, and construct a texture descriptor; Obtain the annotated image of the lung cancer region, perform color space transformation, output a multi-channel image, and send the multi-channel image to a density distribution calculation program to generate color distribution features; Integrate the kernel statistical features and texture descriptors to perform feature combination operations, construct a primary feature group, and send the primary feature group and color distribution features to the feature splicing program to form phenotypic combination features; Import phenotypic combination features to start principal component analysis for feature dimensionality reduction, output the reduced dimensionality features, and perform standardization on the reduced dimensionality features to generate a phenotypic feature set.

5. The artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method according to claim 1, characterized in that: The step of performing hierarchical prediction calculations based on the phenotype feature set to generate immune phenotype prediction results includes: Loading the phenotypic feature set to implement grouping labeling operations, constructing grouping feature data, and sending the grouping feature data to a normalization processing program to output basic prediction features; Obtain basic prediction features to perform correlation analysis and calculation, generate a feature correlation matrix, and initiate threshold filtering on the feature correlation matrix to construct a key feature matrix; Import the key feature matrix and run the subspace decomposition program to form a feature subspace, and then pass the feature subspace into the cross-validation system to establish a validation feature group; Load the verification feature group and perform kernel function transformation to realize feature mapping, output the mapping features, and perform classification score calculation on the mapping features to generate a confidence matrix; The confidence matrix is ​​extracted for probability estimation calculation, the phenotypic probability value is output, and the phenotypic probability value is imported into the confidence threshold screening program to generate the immune phenotype prediction result.

6. The artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method according to claim 1, characterized in that: The method of reading the immune phenotype prediction results and performing subtype mapping analysis to output a patient diagnosis report includes: Import the immune phenotype prediction results to start feature cluster analysis, build immune feature groups, and pass the immune feature groups into the distribution statistics program to generate phenotypic distribution data; Obtain phenotypic distribution data and perform hierarchical clustering operations to make preliminary subtype judgments, form preliminary subtype results, and perform feature contribution analysis on the preliminary subtype results to establish feature importance sequences; Load the feature importance sequence to perform sorting and screening, output the dominant feature set, and send the dominant feature set to the cross-validation system to construct the subtype judgment standard; Extract subtype determination criteria and calculate confidence assessment parameters to generate a subtype confidence matrix, and run an evidence synthesis algorithm on the subtype confidence matrix to form clinical diagnosis data; Collect clinical diagnostic data to perform risk classification operations, establish graded risk indicators, and transfer the graded risk indicators to the quantitative assessment system to generate risk assessment results; Integrate risk assessment results to perform diagnostic recommendation generation processing, output diagnostic recommendation content, and apply a report formatter to the diagnostic recommendation content to construct a patient diagnostic report.

7. The artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method according to claim 1, characterized in that: The receiving of the patient's diagnosis report and performing multi-dimensional data integration processing to generate a personalized treatment plan includes: Load patient diagnostic reports to perform data classification and collation, construct classified diagnostic data, and send the classified diagnostic data to the structured processing program to output structured diagnostic information; Collect structured diagnostic information to initiate feature correlation analysis, form key treatment factors, and apply weight calculation operations to the key treatment factors to generate a treatment factor weight matrix; Import the treatment factor weight matrix to implement hierarchical decomposition processing, establish multi-layer treatment indicators, and pass the multi-layer treatment indicators into the clinical rule matching system to construct a candidate treatment plan set; Obtain a set of candidate treatment options for option evaluation and calculation, output an option scoring matrix, and run a risk-benefit analysis program on the option scoring matrix to form a feasibility evaluation report; Extract the feasibility assessment report to perform the optimal plan calculation, generate a recommended treatment list, and send the recommended treatment list to the drug interaction analysis system to establish a medication safety strategy; Integrate medication safety strategies to implement dose adjustment calculations, construct medication regimens, and initiate scheduling procedures for medication regimens to output treatment schedules; The treatment schedule is loaded for clinical pathway planning and processing to form a treatment execution plan, which is then transferred to the follow-up monitoring design system to generate a personalized treatment plan.

8. An artificial intelligence-based lung cancer phenotype prediction auxiliary analysis system, used to implement the artificial intelligence-based lung cancer phenotype prediction auxiliary analysis method according to any one of claims 1 to 7, characterized in that: The artificial intelligence-based lung cancer phenotype prediction auxiliary analysis system includes: A processing module, used for generating a standardized pre-processed image by performing an adaptive pre-processing operation on the digital image of the lung cancer pathological section; A segmentation module, used for receiving a standardized pre-processed image and performing hierarchical progressive tissue segmentation processing to generate a lung cancer region annotated image; An extraction module is used to read the annotated image of the lung cancer region and perform multi-dimensional feature cascade extraction operations to output a phenotypic feature set; A prediction module, used to perform hierarchical prediction calculations based on the phenotypic feature set to generate immune phenotype prediction results; A mapping module, which is used to read the immunophenotype prediction results and perform subtype mapping analysis to output a patient diagnosis report; The integration module is used to receive patient diagnosis reports and perform multi-dimensional data integration processing to generate personalized treatment plans.

Images