Crohn disease patient treatment evaluation method

Abstract

The invention discloses a treatment evaluation method for a Crohn disease patient, which belongs to the crossing field of medical imaging technology and artificial intelligence, and comprises the following steps: step A, patient image data acquisition and preprocessing: acquiring CTE image data of the Crohn disease patient at a specified time point after treatment, and transmitting the CTE image data to a medical image analysis platform for preprocessing, the preprocessing comprises image denoising and alignment; step B, automatic segmentation and feature extraction of a lesion area: a lesion intestinal wall area VOI is sketched through medical image analysis software, an adjacent mesenteric fat area VOI is automatically constructed, image omics features in the VOI are extracted, a feature matrix is formed, and the image omics features comprise texture features, shape features, intensity features and iodine quantitative features; according to the method, mucous membrane healing evaluation is achieved through CTE examination, invasive operation is not needed, pains and risks of enteroscopy are avoided, a patient can accept the method more easily, and the method is especially suitable for multiple dynamic monitoring after treatment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical imaging technology and artificial intelligence, in particular to a Crohn's disease patient treatment evaluation method. BACKGROUND

[0002] Crohn's disease is a chronic recurrent intestinal inflammatory disease, and the treatment goal is not only to relieve clinical symptoms, but also to achieve mucosal healing, because mucosal healing is a key prognostic indicator for long-term remission, reducing the rate of surgery and hospitalization. At present, the "gold standard" for clinically evaluating mucosal healing of Crohn's disease patients is colonoscopy, which can determine whether healing is achieved by directly observing the lesion state of intestinal mucosa under endoscopy. However, colonoscopy has significant limitations: Colonoscopy requires inserting an endoscope into the intestinal tract, which may cause discomfort, pain, and even risks such as intestinal perforation and bleeding, resulting in poor patient compliance, especially for frequent post-treatment dynamic monitoring. Moreover, the results are greatly influenced by the experience of the operator, and different doctors may have different interpretations of the same lesion, which lacks objectivity. In addition, colonoscopy cannot fully cover the lesions in the deep small intestine and anastomotic stoma, which may miss important lesions.

[0003] To make up for the shortcomings of colonoscopy, computed tomography enterography (CTE) as a non-invasive imaging method has been widely used in the diagnosis and follow-up of Crohn's disease. CTE can clearly show the lesion characteristics of intestinal wall thickening, enhancement, and mesentery changes, providing important information for evaluating intestinal inflammatory activity. However, existing mucosal healing evaluation methods based on CTE mainly rely on subjective interpretation of image features by doctors, such as intestinal wall thickness and enhancement degree, which has low consistency among different observers and is difficult to standardize. Moreover, the features are single, and only focus on the image features of the intestinal wall itself, ignoring the lesion correlation of surrounding tissues such as mesenteric fat. The evaluation accuracy is limited, and the clinical biochemical indicators such as calprotectin and C-reactive protein are not effectively integrated, making it difficult to achieve individualized and accurate evaluation. SUMMARY

[0004] The purpose of the present application is to solve the problems of strong invasiveness of colonoscopy and insufficient accuracy of subjective evaluation of CTE, and to propose a Crohn's disease patient treatment evaluation method.

[0005] To achieve the above purpose, the present application adopts the following technical scheme: A Crohn's disease patient treatment evaluation method, comprising the following steps: Step A: Patient image data acquisition and preprocessing: CTE image data of Crohn's disease patients at a specified time point after treatment is collected and transmitted to a medical image analysis platform for preprocessing, which includes image denoising and alignment. Step B: Automatic segmentation and feature extraction of lesion area: The lesion intestinal wall region VOI is delineated by medical image analysis software, and the adjacent mesenteric fat region VOI is automatically constructed. Radiomic features within the VOI are extracted to form a feature matrix. The radiomic features include texture features, shape features, intensity features and iodine quantification features. Step C: Automatically evaluate the model's operation and MH status output: Input the feature matrix into the fusion radiomics model, which integrates intestinal wall imaging features, mesenteric fat imaging features, and clinical biochemical indicators, and outputs the predicted probability value and risk score of MH; Step D: Clinical decision support and risk visualization: Generate personalized predictive nomograms and standardized reports to assist clinical decision-making.

[0006] Preferably, the CTE image data acquisition parameters include: tube voltage 80-140kVp, tube current automatic adjustment, slice thickness 0.5-1.0mm, pitch 0.5-1.0, and scanning range from the top of the diaphragm to the pubic symphysis.

[0007] Preferably, the medical image analysis software is either ITK-SNAP or nnU-Net.

[0008] Preferably, after feature extraction, a feature optimization step is also included, in which random forest, recursive feature elimination or LASSO regression is used to select features and retain key features related to MH evaluation.

[0009] Preferably, the method for constructing the fusion radiomics model includes: collecting training set data, wherein the training set includes CTE radiomics features and corresponding clinical biochemical indicators and colonoscopy results; training the model using logistic regression, support vector machine or neural network algorithms; and optimizing the model parameters through cross-validation.

[0010] Preferably, the clinical biochemical indicators include calprotectin, C-reactive protein, erythrocyte sedimentation rate, and albumin levels.

[0011] Preferably, the predicted nomogram includes the patient's radiomics score, clinical biochemical index values, and corresponding MH prediction probabilities, and the standardized report is incorporated into the hospital's diagnosis and treatment system.

[0012] Preferably, it also includes a data acquisition module for acquiring CTE imaging data and clinical biochemical indicators; The preprocessing module is used to denoise and align the image data; The segmentation and feature extraction module is used to segment lesion areas and extract radiomics features; The model evaluation module is used to run the fusion radiomics model and output the MH evaluation results. The visualization module is used to generate forecast nomograms and standardized reports.

[0013] Preferably, the segmentation and feature extraction module includes a human-computer interaction unit and an automatic calculation unit. The human-computer interaction unit allows doctors to delineate the lesion area, and the automatic calculation unit automatically constructs the VOI of the mesenteric fat region and extracts features.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Mucosal healing assessment based on CTE examination is non-invasive, avoiding the pain and risks of colonoscopy, making it more acceptable to patients, and especially suitable for multiple dynamic monitoring after treatment.

[0015] 2. By extracting the quantitative features of iodine in spectral CT and multidimensional radiomics features, subjective image interpretation is transformed into objective quantitative indicators, reducing the interference of human factors. Clinical verification shows that the evaluation results of this invention are consistent with the gold standard of colonoscopy by more than 85%, which is significantly higher than the traditional subjective assessment of CTE.

[0016] 3. For the first time, quantitative characteristics of intestinal wall iodine, characteristics of mesenteric fat, and clinical biochemical indicators were integrated to capture multi-system changes in the disease (intestinal inflammation, systemic inflammatory response, and mesenteric lesions), which significantly improved the sensitivity and specificity compared to assessment using single features. Attached Figure Description

[0017] Fig. 1 This is a flowchart of patient impact data collection and preprocessing for a treatment assessment method for Crohn's disease patients proposed in this invention; Fig. 2 This is a schematic diagram of the automatic segmentation and feature extraction process of the lesion area in a treatment assessment method for Crohn's disease patients proposed in this invention. Fig. 3 This is a schematic diagram of the operation and MH status output process of the automatic assessment model of the Crohn's disease patient treatment assessment method proposed in this invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] Example, refer to Figs. 1-3 A treatment assessment method for Crohn's disease patients includes the following steps: Step A: Patient image data acquisition and preprocessing: Acquire CTE image data of Crohn's disease patients at specified time points after treatment, and transmit them to the medical image analysis platform for preprocessing, which includes image denoising and alignment; Step B: Automatic segmentation and feature extraction of lesion area: The lesion intestinal wall region VOI is delineated by medical image analysis software, and the adjacent mesenteric fat region VOI is automatically constructed. Radiomic features within the VOI are extracted to form a feature matrix. The radiomic features include texture features, shape features, intensity features and iodine quantification features. Step C: Automatically evaluate the model's operation and MH status output: Input the feature matrix into the fusion radiomics model, which integrates intestinal wall imaging features, mesenteric fat imaging features, and clinical biochemical indicators, and outputs the predicted probability value and risk score of MH; Step D: Clinical decision support and risk visualization: Generate personalized predictive nomograms and standardized reports to assist clinical decision-making.

[0020] In a preferred embodiment, the CTE image data acquisition parameters include: tube voltage 80-140kVp, tube current automatic adjustment, slice thickness 0.5-1.0mm, pitch 0.5-1.0, and scanning range from the top of the diaphragm to the pubic symphysis.

[0021] In this technical solution: Patient image data acquisition and preprocessing steps: CTE Imaging Data Acquisition: Crohn's disease patients undergo routine CTE examinations at designated time points after treatment, such as 8 weeks and 12 weeks post-treatment. Prior to the examination, patients need to undergo bowel preparation, such as oral administration of a negative contrast agent. The scan is performed using an energy-dispersive CT scanner with the following parameters set: tube voltage 80-140kVp, switchable dual-energy mode, tube current using automatic exposure control technology, slice thickness 0.5-1.0mm, slice spacing 0.5-1.0mm, pitch 0.5-1.0mm, and scanning range from the top of the diaphragm to the pubic symphysis to completely cover the entire intestine. Data transmission and preprocessing: After the scan is completed, the CTE image data, including the raw tomographic images and energy dispersive spectroscopy (EDS) images, is automatically transmitted to a dedicated medical image analysis platform via the hospital's PACS system. This platform uses a standardized interface to call preprocessing algorithms and automatically completes the following operations: Image denoising: Gaussian filtering and nonlocal means denoising algorithms are used to remove noise from the image while preserving lesion details; Image alignment: Spatially aligning multiple examination images of the same patient to eliminate errors caused by differences in scanning position; In a preferred embodiment, refer to Figs. 1-3The medical image analysis software is either ITK-SNAP or nnU-Net.

[0022] In this technical solution: Automatic segmentation and feature extraction steps for lesion areas: VOI delineation in the lesion area: In medical image analysis software, such as ITK-SNAP, doctors manually delineate or modify the intestinal wall region (Volume of Interest, VOI) of the lesion segment through human-computer interaction. Based on the delineated intestinal wall VOI, the software system automatically identifies and constructs the VOI of the adjacent mesenteric fat region, extending outward from the intestinal wall VOI to a range of 5-10 mm of adipose tissue. An alternative is to use automated segmentation tools, such as the nnU-Net deep learning model, to automatically segment the VOI. Doctors only need to fine-tune the segmentation results, improving efficiency.

[0023] Radiomics Feature Extraction: The software system extracts features from the intestinal wall VOI and mesenteric fat VOI, specifically including: Texture features include gray-level co-occurrence matrix features (energy, entropy, contrast, correlation), gray-level run-length matrix features (long run advantage, short run advantage), wavelet transform features, etc., which reflect the uniformity of gray-level distribution and spatial structure of the lesion area. Shape characteristics: including the volume, surface area, sphericity, and standard deviation of wall thickness of the VOI in the intestinal wall, describing the morphological changes of the diseased intestinal segment; Intensity characteristics include average CT value, standard deviation of CT value, maximum CT value, minimum CT value, etc., reflecting the density differences of tissues.

[0024] Feature optimization: The extracted high-dimensional features are reduced in dimensionality using a feature selection algorithm, retaining the key features most relevant to mucosal healing. Algorithms that can be used include: Random Forest algorithm: By calculating feature importance scores, the top 20% of features are selected. Recursive feature elimination: It gradually removes features that contribute the least to the model performance until the optimal feature subset is retained; In a preferred embodiment, refer to Figs. 1-3 After feature extraction, a feature optimization step is also included, which uses random forest, recursive feature elimination or LASSO regression to select features and retain key features related to MH evaluation.

[0025] Furthermore, the method for constructing the fusion radiomics model includes: collecting training set data, wherein the training set includes CTE radiomics features and corresponding clinical biochemical indicators and colonoscopy results; training the model using logistic regression, support vector machine or neural network algorithms; and optimizing the model parameters through cross-validation.

[0026] Furthermore, the clinical biochemical indicators include calprotectin, C-reactive protein, erythrocyte sedimentation rate (ESR), and albumin levels.

[0027] Furthermore, the predicted nomogram includes the patient's radiomics score, clinical biochemical index values, and corresponding MH prediction probabilities, and the standardized report is incorporated into the hospital's diagnosis and treatment system.

[0028] Furthermore, it also includes The data acquisition module is used to collect CTE imaging data and clinical biochemical indicators; The preprocessing module is used to denoise and align the image data; The segmentation and feature extraction module is used to segment lesion areas and extract radiomics features; The model evaluation module is used to run the fusion radiomics model and output the MH evaluation results. The visualization module is used to generate forecast nomograms and standardized reports.

[0029] Furthermore, the segmentation and feature extraction module includes a human-computer interaction unit and an automatic calculation unit. The human-computer interaction unit allows doctors to delineate lesion areas, and the automatic calculation unit automatically constructs the VOI of the mesenteric fat region and extracts features.

[0030] In this technical solution: LASSO regression: It achieves feature selection by compressing the coefficients of irrelevant features to zero through regularization penalty coefficients.

[0031] Automatically evaluate the model's operation and MH state output: Construction of a fusion radiomics model: Training set data: Collect CTE radiomics features, clinical biochemical indicators (calprotectin, C-reactive protein, erythrocyte sedimentation rate, albumin) and colonoscopy results (MH is defined as complete disappearance of intestinal mucosal ulcers under colonoscopy as the gold standard label) of patients with previous Crohn's disease. Model training: Using logistic regression, support vector machine (SVM) or neural network algorithms, with radiomics features and clinical biochemical indicators as inputs and MH states as outputs, a fusion evaluation model is trained. Model optimization: The model hyperparameters (such as the kernel function parameters of SVM and the number of hidden layer nodes of neural network) were adjusted by 5-fold cross-validation, and the optimal diagnostic threshold was determined by receiver operating characteristic (ROC) curve analysis.

[0032] Model evaluation and result output: Input the optimized features of the current patient into the trained fusion model, and the model will run automatically and output the predicted probability value of MH (0-100%). The predicted probability values ​​were converted into risk scores (0-10 points), where a score ≤3 was low risk (MH state), 3-7 was medium risk (partial healing), and ≥7 was high risk (active inflammation). The assessment results are displayed in real time on the interface used by doctors, clearly indicating the patient's current MH status.

[0033] Clinical decision support and risk visualization tools: Personalized prediction nomogram: The software system automatically generates a nomogram, which displays key factors affecting MH (such as calprotectin level and intestinal wall thickness) in the form of coordinate axes. The system intuitively calculates the patient's MH prediction probability through connecting lines and marks the risk level (low, medium, high).

[0034] Standardized report generation: The system automatically integrates basic patient information, CTE imaging features, model prediction results, risk scores and nomograms to generate a standardized assessment report, which is then incorporated into the patient's electronic medical record through the hospital information system (HIS) interface to assist doctors in developing subsequent treatment plans (such as maintaining the original treatment, adjusting drug dosage or changing the treatment plan).

[0035] Dynamic monitoring function: For patients who have undergone multiple follow-up examinations, the system can compare the assessment results at different time points and generate trend curves to intuitively display the dynamic changes in mucosal healing.

[0036] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A treatment assessment method for Crohn's disease patients, characterized in that, Includes the following steps: Step A: Patient image data acquisition and preprocessing: Acquire CTE image data of Crohn's disease patients at specified time points after treatment, and transmit them to the medical image analysis platform for preprocessing, which includes image denoising and alignment; Step B: Automatic segmentation and feature extraction of lesion area: The lesion intestinal wall region VOI is delineated by medical image analysis software, and the adjacent mesenteric fat region VOI is automatically constructed. Radiomic features within the VOI are extracted to form a feature matrix. The radiomic features include texture features, shape features, intensity features and iodine quantification features. Step C: Automatically evaluate the model's operation and MH status output: Input the feature matrix into the fusion radiomics model, which integrates intestinal wall imaging features, mesenteric fat imaging features, and clinical biochemical indicators, and outputs the predicted probability value and risk score of MH; Step D: Clinical decision support and risk visualization: Generate personalized predictive nomograms and standardized reports to assist clinical decision-making.

2. The method for evaluating the treatment of Crohn's disease patients according to claim 1, characterized in that, The CTE image data acquisition parameters include: tube voltage 80-140kVp, tube current automatic adjustment, slice thickness 0.5-1.0mm, pitch 0.5-1.0, and scanning range from the top of the diaphragm to the pubic symphysis.

3. The method for evaluating the treatment of Crohn's disease patients according to claim 1, characterized in that, The medical image analysis software is either ITK-SNAP or nnU-Net.

4. The method for evaluating the treatment of Crohn's disease patients according to claim 1, characterized in that, After feature extraction, a feature optimization step is also included, which uses random forest, recursive feature elimination or LASSO regression to select features and retain key features that are relevant to MH evaluation.

5. The method for evaluating the treatment of Crohn's disease patients according to claim 1, characterized in that, The method for constructing the fusion radiomics model includes: collecting training set data, which includes CTE radiomics features and corresponding clinical biochemical indicators and colonoscopy results; training the model using logistic regression, support vector machine or neural network algorithms; and optimizing the model parameters through cross-validation.

6. The method for evaluating the treatment of Crohn's disease patients according to claim 1, characterized in that, The clinical biochemical indicators include calprotectin, C-reactive protein, erythrocyte sedimentation rate (ESR), and albumin levels.

7. The method for evaluating the treatment of Crohn's disease patients according to claim 1, characterized in that, The predicted nomogram includes the patient's radiomics score, clinical biochemical index values, and corresponding MH prediction probability, and the standardized report is incorporated into the hospital's diagnosis and treatment system.

8. The method for evaluating the treatment of Crohn's disease patients according to claim 1, characterized in that, Also includes The data acquisition module is used to collect CTE imaging data and clinical biochemical indicators; The preprocessing module is used to denoise and align the image data; The segmentation and feature extraction module is used to segment lesion areas and extract radiomics features; The model evaluation module is used to run the fusion radiomics model and output the MH evaluation results. The visualization module is used to generate forecast nomograms and standardized reports.

9. The method for evaluating the treatment of Crohn's disease patients according to claim 8, characterized in that, The segmentation and feature extraction module includes a human-computer interaction unit and an automatic calculation unit. The human-computer interaction unit allows doctors to delineate lesion areas, and the automatic calculation unit automatically constructs the VOI of the mesenteric fat region and extracts features.

Images