Method and system for predicting the curative effect of psoriasis by combining traditional Chinese and western medicine based on RCM images

Abstract

This invention discloses a method and system for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, relating to the fields of medical diagnosis and medical image processing technology. The method includes: acquiring raw digitally stitched RCM images of psoriasis target lesions before treatment; automatically quantifying and extracting multi-dimensional microscopic indicators such as epidermal thickness, density of incomplete keratinization nuclei, density of inflammatory cell infiltration, diameter of vascular loops, and granular layer state based on deep learning algorithms; substituting the standardized indicators into a weighted model iteratively trained from historical data to calculate and output the R-PSI prediction index; comparing it with a preset diagnostic threshold to output a binary judgment result of efficacy. This invention achieves a leap from post-treatment verification to pre-treatment prediction in efficacy evaluation, solving the problems of subjective lag in traditional assessment and invasive and expensive molecular diagnosis, providing non-invasive, objective, and quantitative decision support for personalized precision treatment.

Description

Technical Field

[0001] This invention belongs to the field of medical image data processing technology, specifically relating to a data processing method and system for predicting the efficacy of psoriasis treatment based on RCM images. Background Technology

[0002] Psoriasis is an immune-mediated chronic inflammatory skin disease with a complex pathological mechanism, a prolonged course, and a tendency to relapse. In recent years, treatment options for psoriasis have become increasingly diversified, encompassing traditional drugs, biologics, and traditional Chinese medicine therapies. In clinical practice, the combined use of traditional Chinese medicine formulas (such as Xiaobi Decoction) and biologics (such as IL-17A inhibitors) has shown great clinical application potential as a synergistic and toxicity-reducing combined treatment strategy. However, due to significant immune heterogeneity among individual patients, not all patients can obtain equivalent benefits from this specific integrated traditional Chinese and Western medicine therapy. To achieve personalized and precise treatment, avoid the waste of medical resources and delays in treatment caused by ineffective medication, the field has been exploring how to accurately predict a specific patient's response to a specific therapy before treatment begins.

[0003] Currently, the clinical "gold standard" for assessing the efficacy and severity of psoriasis treatment still primarily relies on the Psoriasis Area and Severity Index (PASI) score. However, this method has significant technical limitations: on the one hand, its assessment process is highly dependent on the physician's visual observation and subjective experience, making it quite subjective; on the other hand, the macroscopic lesion score is extremely insensitive to changes in the deep layers and early microstructures of the lesions, and can only serve as a lagging evaluation indicator "after treatment," completely unsuitable for the prospective prediction of efficacy "before treatment."

[0004] In the field of predictive methodologies, current efficacy predictions primarily rely on molecular diagnostic techniques. For example, by drawing blood from patients or performing skin biopsies, the expression levels of specific genes or single nucleotide polymorphisms (SNPs) in the samples are detected to determine whether a patient is a "high responder" or a "low responder." The core of these methods lies in establishing the association between molecular markers and efficacy, but their practical clinical application faces several bottlenecks: the testing process is invasive, easily increasing patient discomfort and infection risks; the testing costs are high, the analysis process is complex and time-consuming; more importantly, molecular testing strips away the original morphological context of skin lesions, failing to provide intuitive, real-time microscopic histopathological information, making it difficult to widely apply in primary healthcare institutions and outpatient settings requiring rapid decision-making.

[0005] Meanwhile, in the field of non-invasive medical imaging, reflective confocal microscopy (RCM), as an "optical biopsy" technique, has been introduced into the clinical examination of psoriasis. RCM can directly observe changes in microstructures such as epidermal thickness and inflammatory cell infiltration through non-invasive optical scanning, thus objectively reflecting the current pathological state of the skin lesions. However, current applications of RCM technology mainly remain at the level of "static description" and "post-treatment monitoring" of skin lesion characteristics, with a serious lack of depth in its application. How to establish a mathematical mapping relationship between imaging features and future treatment efficacy, and how to transform the massive amounts of microstructured data collected by RCM into "digital biomarkers" with quantitative predictive value, represents a current technological gap in radiomics.

[0006] In summary, the treatment of psoriasis, especially in integrated traditional Chinese and Western medicine treatment plans, urgently needs to overcome the multiple shortcomings of existing clinical scoring methods, such as subjective lag, invasive and expensive molecular diagnostics, and the lack of predictive dimensions in current RCM applications. Currently, the industry severely lacks a method and system capable of non-invasive, objective, rapid, and quantitative prediction of a patient's potential therapeutic effect before treatment begins. In particular, existing technologies lack mathematical mapping models that can automatically quantify and establish prospective predictive value using non-invasive imaging indicators, thus failing to provide targeted "pre-treatment" decision-making references for integrated traditional Chinese and Western medicine therapies. Summary of the Invention

[0007] 1. The technical problem to be solved by the present invention

[0008] In view of the numerous limitations of existing technologies in the treatment of psoriasis, especially in integrated traditional Chinese and Western medicine treatment plans, the purpose of this invention is to provide a method and system for predicting the efficacy of integrated traditional Chinese and Western medicine treatment for psoriasis based on RCM imaging, and to address the following core technical problems existing in the prior art:

[0009] (1) Regarding the lack of specific predictive tools for integrated traditional Chinese and Western medicine therapies, which leads to blind clinical application: Although integrated traditional Chinese and Western medicine therapies (such as traditional Chinese medicine compound combined with IL-17A inhibitors) have shown great potential, there is a lack of effective tools in the current technology that can objectively predict whether a specific patient will benefit from the combined therapy before treatment. This results in clinicians lacking forward-looking basis when making individualized medication decisions, and there is a certain degree of blind treatment.

[0010] (2) Regarding the problems of invasiveness, high cost and difficulty in popularization of molecular diagnostic prediction methods: Existing gene or protein-based molecular prediction pathways usually require blood draws or tissue biopsies, which are obviously invasive, and the detection costs are high, the process is complicated and time-consuming. In addition, such methods cannot provide real-time information on the morphology of skin lesions, making it difficult to widely apply them in primary healthcare institutions and outpatient settings that require rapid decision-making.

[0011] (3) Regarding the insufficient depth of application and inadequate data mining of RCM imaging technology: The existing application of reflective confocal microscopy (RCM) is mainly limited to the static description of skin lesions and dynamic monitoring after treatment, and the potential of the massive amount of microstructured data collected has not been fully explored in the "treatment prediction" level. How to transform these microscopic imaging indicators into digital biomarkers with high predictive value is a technical gap that urgently needs to be filled in the current field of medical imaging omics.

[0012] (4) Addressing the issues of strong subjectivity and lack of quantitative prediction models in existing clinical assessment methods: Traditional assessment standards such as PASI scores are highly subjective and insensitive to subtle pathological changes in the early stages of treatment. This invention aims to overcome the drawbacks of subjectivity by deeply mining the RCM image features of skin lesions before treatment and establishing a quantitative assessment and weighted prediction model, thereby achieving non-invasive, objective, rapid, and quantitative accurate prediction of the efficacy of integrated traditional Chinese and Western medicine.

[0013] 2. Technical Solution

[0014] To address the aforementioned technical problems, this invention provides a method, system, and application for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, employing the following technical solution:

[0015] The first aspect of this invention provides a method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, characterized by comprising the following steps:

[0016] S1. RCM image standardization acquisition: Obtain raw digital stitched images of the RCM of the target skin lesion area of ​​psoriasis before treatment;

[0017] S2. Automated extraction and quantification of microscopic features: The original digitally stitched image of the RCM is processed to automatically quantify and extract microscopic indicators that characterize the pathological state of psoriasis. The microscopic indicators include epidermal thickness ET, density of incomplete keratinization nuclei DPC, density of inflammatory cells infiltrating the dermal papillae IIDP, average diameter of dermal papillary vascular loops MDVP, and granular layer status GCS score.

[0018] S3. Construct and calculate the R-PSI prediction index: Standardize the quantified micro-indices and substitute them into the preset weighted model to calculate and output the R-PSI prediction index.

[0019] S4. Data classification output based on R-PSI threshold: The calculated R-PSI prediction index is converted into a probability value and compared with a preset classification threshold to output data classification labels and probability feature values ​​used to characterize the trend of changes in the microstructure of skin lesions.

[0020] Furthermore, the specific process of acquiring the original digitally stitched RCM image in step S1 includes: controlling the reflective confocal microscope to irradiate with a laser wavelength of 830 nm and a surface laser power of less than 7.5 mW; using a scanning method combining Z-Stack mode and Mosaic mode, scanning from the surface of the stratum corneum to the papillary dermis at a scanning frame rate of 12 frames / second, with a step interval of 3-5 μm, and the scanning depth extending to the upper part of the papillary dermis; stitching the acquired field-of-view images to generate the original digitally stitched RCM image with a size of 3 mm × 3 mm.

[0021] Furthermore, before extracting the microscopic indicators in step S2, a skin layering step based on deep learning is also included: using a U-Net structured image segmentation network to perform pixel-level segmentation on the original digitally stitched RCM image, automatically identifying the epidermal layer region, the dermal-epidermal junction DEJ boundary, the dermal papillary layer region, and the granular layer region located at the bottom of the epidermis and above the DEJ.

[0022] Further, the specific quantification logic and feature engineering algorithm for extracting various micro-indicators in step S2 include: obtaining the ET: calculating the shortest Euclidean distance from each pixel in the epidermal region to the DEJ boundary, and calculating the average of all vertical distances; obtaining the DPC density: in the segmented stratum corneum region, applying Top-hat transform to enhance the bright small-sized nucleus targets, using the Otsu adaptive thresholding method to binarize the enhanced image, and using connected component analysis operators to filter connected components whose area and roundness features conform to the preset cell nucleus model, and counting the total number of connected components per unit area; obtaining the IIDP: in the segmented... Within the dermal papillary layer region, the Laplacian-Gaussian operator is used as the speckle detection operator to locate highly refractive bright specks, and the total number of bright specks detected per unit area is counted; the MDVP is obtained: using a U-Net segmentation network specifically trained for blood vessel recognition, the contours of blood vessel loops within the dermal papillary layer region are accurately delineated, the equivalent circle diameter of each segmented blood vessel loop section is calculated, and the average value of the diameters of all measured blood vessel loops within the field of view is counted; the GCS score is obtained: the contrast and energy texture feature parameters of the granular layer region are calculated using the gray-level co-occurrence matrix algorithm, and the calculated texture feature parameters are mapped to a standardized GCS score.

[0023] Furthermore, the specific solution process in step S3 includes: using the Z-score normalization algorithm to transform the original values ​​of each extracted micro-indicator into a normal distribution. ,in These are the original values. and These are the statistical mean and standard deviation of the corresponding indicator in the historical training dataset; the R-PSI prediction index is calculated using a linear combination formula: Wherein, the preset weighting coefficient The values ​​of the preset weight coefficients are calculated by iteratively training historical clinical case data using a logistic regression model with L2 regularization and a log loss function, and the range of values ​​for each preset weight coefficient is limited as follows: , , , , .

[0024] Furthermore, the decision-making process for binarying the efficacy assessment in step S4 includes: converting the calculated R-PSI predictive index into a probability value between 0 and 1 using a Sigmoid function. The trained model is analyzed using the Receiver Operating Characteristic (ROC) curve, and the Youden index is calculated for each point on the curve. The R-PSI value corresponding to the maximum value of the Youden index is used as the preset diagnostic threshold to distinguish between high-response and low-response populations. When the R-PSI predictive index of a new patient is greater than or equal to the preset diagnostic threshold, the system determines that the patient is a high-probability responder.

[0025] A second aspect of the present invention provides a system for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, used to perform the method described in the first aspect. The system includes: an RCM image acquisition module for acquiring raw digitally stitched RCM images of the target skin lesion area before treatment begins; an image processing and quantification module for automatically extracting and quantifying multi-dimensional microscopic indicators characterizing the pathological state of psoriasis; and an R-PSI calculation and prediction module for calculating the R-PSI prediction index and outputting the final efficacy prediction result in combination with a preset diagnostic threshold.

[0026] Furthermore, the preset diagnostic threshold is 0.68; the system also includes: a data storage module, which is communicatively connected to the R-PSI calculation and prediction module, for storing historical clinical case data, preset weight coefficients obtained from training, and records of patient efficacy prediction results; and a display module, for presenting the predicted high response probability value, the quantitative values ​​of multi-dimensional microscopic indicators, and the corresponding RCM pathological images to the clinical terminal.

[0027] A third aspect of the present invention provides the application of the method as described in the first aspect, or the system as described in the second aspect, or the calculated R-PSI predictive index, in the preparation or construction of efficacy assessment tools to guide individualized treatment plans for psoriasis; particularly in the construction of an auxiliary decision-making model for assessing whether to use the specific therapy of "Xiaobi Decoction combined with IL-17A inhibitor".

[0028] It should be noted that the R-PSI prediction index and probability value output by the method provided in this invention are only intermediate results of computer processing of image data, and are intended to provide objective and quantitative data reference for clinical decision-making, not for obtaining specific disease diagnosis results or directly formulating medical plans.

[0029] 3. Beneficial effects

[0030] (1) This invention overcomes the lag in traditional assessments, enabling a methodological shift in psoriasis efficacy evaluation from "post-treatment" to "pre-treatment." Existing clinical scores (such as PASI) and routine RCM applications primarily focus on post-treatment validation and status monitoring, failing to provide prospective guidance before treatment. This invention proposes and implements for the first time a radiomics-based efficacy prediction mechanism. By deeply mining the microstructured data of pre-treatment skin lesions, a mathematical mapping relationship between imaging indicators and efficacy outcomes is established. This approach fills the technological gap in the field of integrated traditional Chinese and Western medicine treatment, which lacks early specific predictive tools, empowering clinicians with crucial pre-treatment decision-making capabilities.

[0031] (2) Based on a specific safe photoelectric parameter acquisition mechanism, a completely non-invasive and highly safe surface detection solution is provided. Unlike invasive molecular diagnostic techniques that rely on blood tests or skin tissue biopsies, this invention strictly limits the 830 nm near-infrared laser wavelength and the surface laser power to below 7.5 mW in its technical solution. This acquisition mechanism only requires optical scanning of the target skin lesion to obtain deep tissue information, achieving a painless and non-invasive process throughout, effectively avoiding the infection risk brought by invasive examinations, and is extremely safe, especially suitable for long-term, high-frequency repeated monitoring of patients with chronic inflammatory skin diseases.

[0032] (3) An objective quantitative system based on AI algorithms was constructed, eliminating the subjective bias of manual assessment from the ground up. Addressing the drawback of traditional manual scoring being easily influenced by doctors' subjective experience, this invention establishes objective quantitative standards by deeply integrating computer vision algorithms during the feature extraction stage. Specifically, the U-Net deep learning network is used to accurately segment the anatomical layers of the skin. Combined with Top-hat transform, Laplacian-Gaussian (LoG) operator, and gray-level co-occurrence matrix (GLCM), pixel-level automatic quantification of epidermal thickness (ET), incomplete keratinization nucleus (DPC), inflammatory cell infiltration (IIDP), vascular loop diameter (MDVP), and granular layer state (GCS) is achieved without manual intervention. This standardized feature engineering ensures a high degree of consistency and repeatability of the assessment results.

[0033] (4) The introduction of a highly robust machine learning model and dynamic threshold determination significantly improves the accuracy and cost-effectiveness of clinical decision-making. This invention does not simply list indicators, but iteratively trains each micro-indicator using a Logistic regression model with L2 regularization and a logarithmic loss function, assigning scientific feature weights to each indicator. By combining the maximum value of the Youden index (preset threshold 0.68) extracted from the receiver operating characteristic (ROC) curve analysis, the system can accurately screen out "potential high responders" and "potential low responders". This not only helps doctors develop more targeted individualized treatment plans, but also effectively avoids patients blindly using expensive and potentially ineffective biological agent combination therapies, greatly reducing the economic burden on patients while significantly saving and optimizing the allocation of medical resources.

[0034] (5) It achieves automation and rapid response throughout the entire diagnosis and treatment process, and has excellent prospects for clinical promotion and translation. The system architecture of this invention is logically rigorous, and the automated image processing and the calculation of the R-PSI prediction index are both completed instantaneously by the processing unit. Within minutes after the patient completes the RCM examination, the predicted probability and visualized pathological images can be output immediately, which greatly shortens the time cycle of clinical decision-making. This system can be seamlessly integrated into existing RCM equipment workstations, or it can be deployed as an independent cloud analysis service, making it very easy to widely promote in medical institutions at all levels and outpatient settings. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a flowchart illustrating a method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, as provided in an embodiment of the present invention.

[0037] Figure 2 A modular structure diagram of a system for predicting the efficacy of integrated traditional Chinese and Western medicine in psoriasis based on RCM imaging, provided for an embodiment of the present invention;

[0038] Figure 3 A flowchart illustrating the algorithm logic of the R-PSI model during the training phase, as provided in this embodiment of the invention.

[0039] Figure 4 A flowchart illustrating the prediction algorithm logic of the R-PSI model during the inference phase, as provided in this embodiment of the invention.

[0040] Figure 5This is a schematic diagram of RCM microscopic pathological feature extraction provided in an embodiment of the present invention;

[0041] Figure 6 The receiver operating characteristic (ROC) curve of the prediction model provided in the embodiments of the present invention;

[0042] Figure 7 Box plots showing the distribution of R-PSI prediction index for high-response and low-response groups provided in embodiments of the present invention.

[0043] The module includes: 10-RCM image acquisition module; 20-image processing and quantization module; 30-R-PSI calculation and prediction module; 40-data storage module; and 50-display module. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0045] Example 1: A method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging.

[0046] like Figure 1 The diagram shown is a flowchart illustrating a method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, as provided in this embodiment. This method is primarily used to predict the potential efficacy for patients with moderate to severe psoriasis vulgaris before initiating integrated traditional Chinese and Western medicine treatments such as "Xiaobi Decoction combined with IL-17A inhibitors." Specifically, it includes the following steps:

[0047] S1. RCM Image Standardization Acquisition: Before treatment, a lesion with typical erythema, infiltration, and scaling characteristics on the extensor surface of the patient's trunk or limbs, which is untreated, is selected as the target lesion area. An in vivo optical scan of the target lesion area is performed using a reflective confocal microscope (RCM). To ensure the safety of the acquisition and the consistency of image quality, the acquisition parameters of the RCM equipment are strictly controlled: irradiation is performed using an 830 nm near-infrared laser wavelength and a surface laser power of less than 7.5 mW; a scanning method combining Z-Stack and Mosaic modes is used, with a scanning frame rate of 12 frames / second, performing a step-by-step scan from the stratum corneum surface (depth 0 μm) towards the papillary dermis, with a step interval set to 3-5 μm, extending the scanning depth to the upper part of the papillary dermis (approximately 250 μm); finally, the acquired single-frame field-of-view images are stitched together to generate a 3mm × 3mm original digital stitched image of the RCM, which serves as input data for subsequent feature extraction.

[0048] S2. Automated extraction and quantification of microscopic features: such as Figure 5 The diagram shown illustrates the extraction of microscopic pathological features from the RCM (Reptilia Craniotomy) image according to an embodiment of the present invention. This step involves deep learning-based image processing of the original digitally stitched RCM image to automatically quantify and extract five microscopic indicators characterizing the pathological state of psoriasis:

[0049] First, a U-Net image segmentation network is used to perform pixel-level segmentation on the original digitally stitched RCM images. The U-Net image segmentation network is a deep learning model pre-trained on a historical RCM image dataset (e.g., containing no fewer than 3000 labeled images) with expert-annotated labels for the epidermal-epidermal junction (DEJ) boundary and granular layer, until the model's loss function converges. Due to its unique U-shaped structure and skip connection design, the U-Net network can effectively combine deep semantic information and shallow spatial information, thereby automatically and accurately identifying the epidermal region, the dermal-epidermal junction (DEJ) boundary, the dermal papillary layer region, and the granular layer region. Subsequently, specific feature extraction is performed:

[0050] (1) Epidermal thickness (ET) and density of incompletely keratinized nuclei (DPC): such as Figure 5 Part A of the diagram illustrates the characteristics of epidermal thickness and incompletely keratinized nuclei. When acquiring the ET (Earthquake Transform), the algorithm calculates the vertical distance from each pixel within the epidermal region to the nearest DEJ (Deep Electrode Layer) boundary and calculates the average value. When acquiring the DPC (Diagram of Dendritic Polymorphic Cells), within the stratum corneum region, a Top-hat transform is applied to enhance the brightened small-sized nuclei. The enhanced image is binarized using the Otsu adaptive thresholding method. Connected component analysis operators are used to filter connected components whose area and roundness features conform to a preset cell nucleus model, and the total number of DPCs per unit area is counted.

[0051] (2) Intracranial cell infiltration density (IIDP) and mean diameter of dermal papillary vascular loops (MDVP): e.g. Figure 5 Part B of the diagram illustrates the characteristics of inflammatory cells and vascular loops within the dermal papilla. When acquiring the IIDP (Intense Inflammatory Diaphragm Species), the Laplacian-Gaussian (LoG) operator was used as the speckle detection operator within the dermal papillary layer region to accurately locate highly refractive, bright inflammatory cell specks and count the total number of specks per unit area. When acquiring the MDVP (Massively Depressed Variants Species Species), a specially trained U-Net segmentation network was used to outline the vascular loop contours, calculate the equivalent circle diameter of each vascular loop cross-section, and calculate the average value.

[0052] (3) Granular layer state (GCS) score: such as Figure 5Part C of the diagram illustrates the texture analysis features of the granular layer. The contrast and energy texture feature parameters of the granular layer region are calculated using the Gray-Level Co-occurrence Matrix (GLCM) algorithm. Normal granular layers typically have high energy values ​​and low contrast; while the blurred granular layers in psoriatic lesions have low energy values ​​and high contrast. The algorithm maps the calculated texture feature parameters to a standardized GCS score. Specifically, the mapping logic for obtaining the GCS score includes: pre-statistically calculating the extreme contrast and energy values ​​of normal skin samples and typical psoriatic lesion samples under the GLCM; mapping the texture feature parameters of the sample to be tested to the standardized interval [0,1] using a linear normalization function, where 1 represents the most complete granular layer structure and 0 represents the complete disappearance of the granular layer.

[0053] S3. Construct and calculate the R-PSI predictive index: This step involves both the model training phase and the inference prediction phase for new patients. During the model training phase (e.g....), Figure 3 As shown): Historical clinical case data was obtained (e.g., data from 120 patients treated with Xiaobi Decoction combined with IL-17A inhibitors, using PASI 90 achievement at 12 weeks as the gold standard for high / low response). When constructing the L2 regularized logistic regression model, the ratio of "high response" to "low response" samples extracted from the historical case database was set to 1:1 to ensure the fairness of the classification boundary and the predictive robustness of the model. The Z-score normalization algorithm was used to transform the original values ​​of the extracted micro-indicators into a normal distribution, as shown in the formula: Subsequently, a Logistic Regression model with L2 regularization was constructed and iteratively trained using a log loss function, outputting optimized preset weight coefficients. After training, the range of values ​​for the weight coefficients corresponding to each micro-indicator is limited to: , , , , In the reasoning stage (such as...) Figure 4 (As shown): Obtain the raw values ​​of various micro-indicators of new patients and use the mean obtained from the training set. and standard deviation Perform Z-score standardization and substitute the results into the linear combination formula to calculate the R-PSI predictive index: .

[0054] S4. Efficacy prediction based on R-PSI threshold: The calculated R-PSI predictive index is converted into a probability value between 0 and 1 using a Sigmoid function. .like Figure 6The figure shows the receiver operating characteristic (ROC) curve of the predictive model provided in this embodiment of the invention. ROC curve analysis of the trained model showed an area under the curve (AUC) of 0.88 (95% CI: 0.82-0.94), indicating excellent discriminative power. By calculating the maximum Youden index at each point on the curve, a preset diagnostic threshold of 0.68 was determined to distinguish between high-responder and low-responder populations. At this threshold, the model's sensitivity reached 92%, and its specificity was 85%. Figure 7 The image shows a box plot of the R-PSI prediction index distribution for the high-response group and the low-response group provided in an embodiment of the present invention. From... Figure 7 It is clear that the vast majority of patients in the high-response group (reflecting 92% sensitivity) had R-PSI values ​​significantly higher than the 0.68 threshold (dashed line); while the vast majority of patients in the low-response group (reflecting 85% specificity) had R-PSI values ​​controlled below the 0.68 threshold (dashed line), with only a very small number of false positive samples exceeding the threshold within the statistically reasonable margin of error. In actual clinical application, when the R-PSI predictive index of a new patient... When the system outputs a binary assessment result for therapeutic efficacy, it indicates "high probability of therapeutic response"; when... At that time, they were judged as "those with a low probability of therapeutic response".

[0055] Example 2: A system for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging.

[0056] like Figure 2 The diagram shown is a block diagram of the module structure of a system for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, according to an embodiment of the present invention. This system is used to execute the prediction method described in Embodiment 1, including:

[0057] 10-RCM Image Acquisition Module: Includes a reflective confocal microscope hardware device, used to acquire raw digitally stitched RCM images of the target lesion area before treatment begins;

[0058] 20-Image Processing and Quantization Module: Integrates deep learning networks (such as U-Net) and feature engineering algorithms (such as Top-hat, GLCM) to automatically extract and quantify multi-dimensional micro-indicators (ET, DPC, IIDP, MDVP, GCS) that characterize the pathological state of psoriasis.

[0059] 30-R-PSI Calculation and Prediction Module: It has a pre-set weighted model and weight coefficients trained with historical data, which are used to standardize the quantified indicators, calculate the R-PSI prediction index, and after converting the probability through the Sigmoid function, it outputs the final binary prediction result of the efficacy in combination with the preset diagnostic threshold (0.68).

[0060] 40-Data Storage Module: Communicates with the R-PSI calculation and prediction module to securely store historical clinical case data, preset weight coefficients obtained from training, and records of patient efficacy prediction results, supporting continuous iteration and updates of the model;

[0061] 50-Display Module: Used to intuitively present the predicted high response probability value, the quantitative values ​​of multi-dimensional microscopic indicators, binary decision suggestions, and the corresponding RCM pathological images (e.g., ...) to the clinical terminal interface. Figure 5 As shown in the figure, it provides clinicians with forward-looking medication decision support.

[0062] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, characterized in that, Includes the following steps: S1. RCM image standardization acquisition: Obtain raw digital stitched images of the RCM of the target skin lesion area of ​​psoriasis before treatment; S2. Automated extraction and quantification of microscopic features: The original digitally stitched image of the RCM is processed to automatically quantify and extract microscopic indicators that characterize the pathological state of psoriasis. The microscopic indicators include epidermal thickness ET, density of incomplete keratinization nuclei DPC, density of inflammatory cells infiltrating the dermal papillae IIDP, average diameter of dermal papillary vascular loops MDVP, and granular layer status GCS score. S3. Construct and calculate the R-PSI prediction index: Standardize the quantified micro-indices and substitute them into the preset weighted model to calculate and output the R-PSI prediction index. S4. Data classification output based on R-PSI threshold: The calculated R-PSI prediction index is converted into a probability value and compared with a preset classification threshold to output data classification labels and probability feature values ​​used to characterize the trend of changes in the microstructure of skin lesions.

2. The method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging according to claim 1, characterized in that, The specific process of acquiring the original digitally stitched image of RCM in step S1 includes: controlling the reflective confocal microscope to irradiate with a laser wavelength of 830nm and a surface laser power of less than 7.5 mW; adopting a scanning method combining Z-Stack mode and Mosaic mode, scanning from the surface of the stratum corneum to the papillary dermis at a scanning frame rate of 12 frames / second, with a step interval of 3-5μm, and the scanning depth extending to the upper part of the papillary dermis; stitching the acquired field of view images to generate the original digitally stitched image of RCM with a size of 3mm×3mm.

3. The method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging according to claim 2, characterized in that, Before extracting the microscopic indicators in step S2, a skin layering step based on deep learning is also included: using a U-Net structured image segmentation network to perform pixel-level segmentation on the original digitally stitched RCM image, automatically identifying the epidermal layer region, the DEJ boundary at the dermal-epidermal junction, the dermal papillary layer region, and the granular layer region located at the bottom of the epidermis and above the DEJ.

4. The method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging according to claim 3, characterized in that, The quantization logic for extracting the ET, MDVP, and GCS score in step S2 is as follows: Obtain the ET: Calculate the shortest Euclidean distance from each pixel in the epidermal region to the DEJ boundary, and calculate the average of all vertical distances; Obtain the MDVP: Using a U-Net segmentation network specifically trained for blood vessel recognition, accurately delineate the vascular loop contours in the dermal papillary layer region, calculate the equivalent circle diameter for each segmented vascular loop cross-section, and calculate the average of all measured vascular loop diameters within the field of view; Obtain the GCS score: Calculate the contrast and energy texture feature parameters of the granular layer region using the gray-level co-occurrence matrix algorithm, and map the calculated texture feature parameters to a standardized GCS score.

5. The method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging according to claim 4, characterized in that, The specific feature engineering algorithm for extracting the DPC density and the IIDP in step S2 includes: obtaining the DPC density: within the segmented stratum corneum region, applying Top-hat transform to enhance the bright small-sized nucleus targets, using the Otsu adaptive thresholding method to binarize the enhanced image, and using connected component analysis operators to filter connected components whose area and roundness features conform to the preset cell nucleus model, and counting the total number of connected components per unit area; obtaining the IIDP: within the segmented dermal papillary layer region, using the Laplacian-Gaussian operator as a speckle detection operator to locate high-refractive-index bright speckles, and counting the total number of bright speckles detected per unit area.

6. The method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging according to claim 1, characterized in that, The specific solution process in step S3 includes: using the Z-score normalization algorithm to transform the original values ​​of each extracted micro-indicator into a normal distribution. ,in These are the original values. and These are the statistical mean and standard deviation of the corresponding indicator in the historical training dataset; the R-PSI prediction index is calculated using a linear combination formula: ;in, These are the preset weighting coefficients for each indicator.

7. The method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging according to claim 6, characterized in that, The preset weighting coefficient The model was calculated by iteratively training historical clinical case data using a Logistic regression model with L2 regularization and a log loss function.

8. The method for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging according to claim 1, characterized in that, The decision-making process for binarying the efficacy assessment in step S4 includes: converting the calculated R-PSI predictive index into a probability value between 0 and 1 using a Sigmoid function. The trained model is analyzed using the Receiver Operating Characteristic (ROC) curve. The Youden index is calculated for each point on the curve. The R-PSI value corresponding to the maximum value of the Youden index is used as the preset diagnostic threshold to distinguish between high-response and low-response populations. When the R-PSI prediction index of a new patient is greater than or equal to the preset diagnostic threshold, the system marks the RCM image data as the first feature category (or high-response feature group).

9. A system for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging, characterized in that, The system for performing the method as described in any one of claims 1 to 8 comprises: an RCM image acquisition module for acquiring raw digitally stitched RCM images of the target lesion area before treatment begins; an image processing and quantification module for automatically extracting and quantifying multi-dimensional microscopic indicators characterizing the pathological state of psoriasis; and an R-PSI calculation and prediction module for calculating the R-PSI prediction index, converting the R-PSI prediction index into a probability value through a Sigmoid function, and then combining it with a preset diagnostic threshold to output the final efficacy prediction result.

10. The system for predicting the efficacy of integrated traditional Chinese and Western medicine in treating psoriasis based on RCM imaging according to claim 9, characterized in that, The preset diagnostic threshold is the R-PSI prediction index corresponding to the maximum value of the Youden index determined by receiver operating characteristic curve analysis; the system also includes: a data storage module, which is communicatively connected to the R-PSI calculation and prediction module, for storing historical clinical case data, preset weight coefficients obtained from training, and records of patient efficacy prediction results; and a display module, which is used to present the predicted high response probability value, the quantitative values ​​of multi-dimensional micro-indicators, and the corresponding RCM pathological images to the clinical terminal.

Images