Hepatocellular carcinoma recurrence condition prediction method and system based on combination of MRI image and clinical features

Abstract

The invention provides a hepatocellular carcinoma recurrence condition prediction method and system combining an MRI image and clinical features. The method comprises the following steps: collecting multi-modal sample MRI image data and preset clinical feature data; performing denoising, normalization processing and size adjustment on the multi-modal sample MRI image data to obtain an adjusted image, and performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image; performing missing value filling and one-hot coding on the clinical feature data to obtain preprocessed feature data; respectively carrying out feature extraction on the enhanced image and the preprocessed feature data to obtain image features and clinical features; fusing the image features and the clinical features, and inputting the fused features into an initial multi-modal deep learning model for training to obtain a hepatocellular carcinoma recurrence condition prediction model; and inputting a to-be-detected MRI image and the clinical features into the hepatocellular carcinoma recurrence condition prediction model to obtain a prediction result. According to the invention, the prediction accuracy and reliability can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hepatocellular carcinoma recurrence prediction, and in particular to a hepatocellular carcinoma recurrence prediction method and system combining MRI images and clinical characteristics. Background Art

[0002] Primary liver cancer is one of the common malignant tumors with high morbidity and mortality. Hepatocellular carcinoma is the most common type of primary liver cancer. Liver resection is the preferred treatment option for patients with hepatocellular carcinoma, but most patients with hepatocellular carcinoma are unable to undergo surgery due to background cirrhosis, liver decompensation, portal hypertension, and insufficient residual liver volume.

[0003] Radiofrequency ablation has the advantages of minimally invasive and few complications, but the early local recurrence rate and overall recurrence rate of patients with hepatocellular carcinoma after radiofrequency ablation are still high. Accurately predicting the risk of early recurrence in patients with hepatocellular carcinoma before radiofrequency ablation can provide a reference for clinicians to formulate treatment plans.

[0004] The invention patent with application number 202310743593.X discloses a method for predicting the recurrence of hepatobiliary carcinoma after surgery, including S1 constructing a recurrence prediction model, the recurrence prediction model includes an image segmentation module, an indicator data analysis module and a prediction analysis module; S2 obtaining a training data set, the training data includes CT images and case indicators; S3 importing the training data set into the recurrence prediction model and optimizing the training; S4 obtaining the data to be predicted; S5 using the optimized recurrence prediction model to analyze the data to be predicted to obtain the recurrence of hepatobiliary carcinoma after surgery; in addition to making full use of CT images, the patent also integrates the indicator data corresponding to the case with its image data to supplement the information that cannot be expressed by the CT image. Through continuous optimization and training, a postoperative recurrence prediction model with good performance was finally obtained. This method relies on a large amount of CT images and case indicator data. The quality and quantity of the data directly affect the performance of the model, resulting in poor prediction results. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a method and system for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical characteristics, which not only improves the accuracy and reliability of the prediction, but also provides strong support for personalized medicine, and has significant clinical application value and research significance.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features, comprising:

[0008] Collect multimodal sample MRI image data and preset clinical feature data of patients with hepatocellular carcinoma;

[0009] Denoising, normalizing and resizing the multimodal sample MRI image data to obtain an adjusted image, and performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image;

[0010] Performing missing value filling and one-hot encoding on the clinical characteristic data to obtain preprocessed characteristic data;

[0011] Performing feature extraction on the enhanced image and the preprocessed feature data respectively to obtain image features and clinical features;

[0012] fusing the image features and the clinical features to obtain fused features;

[0013] Inputting the fusion features into an initial multimodal deep learning model for training to obtain a hepatocellular carcinoma recurrence prediction model;

[0014] The MRI images and clinical features to be tested are input into the hepatocellular carcinoma recurrence prediction model to obtain the prediction results.

[0015] Preferably, the multimodal sample MRI image data includes: T1-weighted imaging data, T2-weighted imaging data and diffusion-weighted imaging data.

[0016] Preferably, the clinical characteristic data include age, gender, liver function index, tumor size, degree of differentiation and AFP level.

[0017] Preferably, performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image comprises:

[0018] Performing filtering on the adjusted image to obtain a filtered image;

[0019] Construct an image enhancement window, and use the image enhancement window to traverse the filtered image for denoising, to obtain a denoised image;

[0020] The denoised image is subjected to histogram equalization processing to obtain an enhanced image.

[0021] Preferably, filtering the adjusted image to obtain a filtered image comprises:

[0022] Performing wavelet decomposition on the adjusted image at multiple scales to obtain corresponding wavelet coefficients;

[0023] constructing a filtering threshold according to the size and decomposition scale of the adjusted image;

[0024] constructing an image filtering model using the filtering threshold;

[0025] The adjusted image is filtered using the image filtering model to obtain the filtered image.

[0026] Preferably, the calculation formula of the filtering threshold is: Among them, λ j represents the filtering threshold at the jth decomposition scale, σ j represents the standard deviation of the noise at the jth decomposition scale, N j represents the signal length at the jth decomposition scale, and j represents the decomposition scale.

[0027] Preferably, the calculation formula of the image filtering model is:

[0028] Using the formula:

[0029]

[0030] Wherein, sign is the sign function, a is the first preset coefficient, b is the second preset coefficient, ω j,k represents the kth wavelet coefficient at the jth decomposition scale, Represents the wavelet coefficients after filtering.

[0031] Preferably, the data enhancement includes rotation, scaling, flipping and random cropping.

[0032] Preferably, the clinical characteristics include: preoperative alpha-fetoprotein and platelet count.

[0033] A hepatocellular carcinoma recurrence prediction system combining MRI images and clinical features, comprising:

[0034] A data collection module, used to collect multimodal sample MRI image data and preset clinical feature data of patients with hepatocellular carcinoma;

[0035] An image data processing module, used for performing denoising, normalization and size adjustment on the multimodal sample MRI image data to obtain an adjusted image, and performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image;

[0036] A clinical data processing module, used for performing missing value filling and one-hot encoding on the clinical characteristic data to obtain pre-processed characteristic data;

[0037] A feature extraction module, used to extract features from the enhanced image and the preprocessed feature data respectively to obtain image features and clinical features;

[0038] A feature fusion module, used for fusing the image feature and the clinical feature to obtain a fusion feature;

[0039] A training module, used for inputting the fusion features into an initial multimodal deep learning model for training to obtain a hepatocellular carcinoma recurrence prediction model;

[0040] The prediction module is used to input the MRI image and clinical characteristics to be tested into the hepatocellular carcinoma recurrence prediction model to obtain the prediction result.

[0041] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0042] The present invention provides a method and system for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features, the method comprising: collecting multimodal sample MRI image data and preset clinical feature data of hepatocellular carcinoma patients; performing denoising, normalization and size adjustment on the multimodal sample MRI image data to obtain an adjusted image, and performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image; performing missing value filling and unique hot encoding on the clinical feature data to obtain preprocessed feature data; performing feature extraction on the enhanced image and the preprocessed feature data respectively to obtain image features and clinical features; fusing the image features and the clinical features to obtain fusion features; inputting the fusion features into an initial multimodal deep learning model for training to obtain a hepatocellular carcinoma recurrence prediction model; inputting the MRI image and clinical features to be tested into the hepatocellular carcinoma recurrence prediction model to obtain a prediction result. The present invention not only improves the accuracy and reliability of prediction, but also provides strong support for personalized medicine, and has significant clinical application value and research significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 A schematic diagram of a method flow chart provided by an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The purpose of the present invention is to provide a method and system for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical characteristics, which not only improves the accuracy and reliability of the prediction, but also provides strong support for personalized medicine, and has significant clinical application value and research significance.

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Figure 1 A schematic diagram of a method flow chart provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features, comprising:

[0050] Step 100: Collect multimodal sample MRI image data and preset clinical feature data of patients with hepatocellular carcinoma;

[0051] Step 200: performing denoising, normalization and resizing on the multimodal sample MRI image data to obtain an adjusted image, and performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image;

[0052] Step 300: performing missing value filling and one-hot encoding on the clinical feature data to obtain preprocessed feature data;

[0053] Step 400: extracting features from the enhanced image and preprocessed feature data respectively to obtain image features and clinical features;

[0054] Step 500: fusing the image features and clinical features to obtain fused features;

[0055] Step 600: inputting the fusion features into the initial multimodal deep learning model for training to obtain a hepatocellular carcinoma recurrence prediction model;

[0056] Step 700: Input the MRI image to be tested and the clinical features into the hepatocellular carcinoma recurrence prediction model to obtain a prediction result.

[0057] Preferably, the multimodal sample MRI image data includes: T1-weighted imaging data, T2-weighted imaging data and diffusion-weighted imaging data.

[0058] Preferably, the clinical characteristic data include age, gender, liver function index, tumor size, degree of differentiation and AFP level.

[0059] Specifically, in this embodiment, the inclusion criteria for the data of patients with hepatocellular carcinoma who underwent radiofrequency ablation treatment include:

[0060] (1) Hepatocellular carcinoma is a single nodule with a long diameter <5 cm, or multiple nodules with a long diameter of <3 cm;

[0061] (2) first treatment, and the treatment method was radiofrequency ablation;

[0062] (3) Complete clinical data;

[0063] (4) Preoperative liver function Child-Pugh grade A or B. Patients with hepatocellular carcinoma with a history of previous treatment, metastasis, or other tumors were excluded.

[0064] In this example, a total of 169 patients with hepatocellular carcinoma were included, including 152 males and 17 females, aged (57.2±9.2) years. Among the 169 patients, 61 had early recurrence. The included patients were randomly divided into a training set (n=135) and a test set (n=34) at a ratio of 8:2. There were 49 recurrences in the training set and 12 recurrences in the test set.

[0065] Furthermore, the observation indicators of this embodiment include:

[0066] (1) preoperative history of hepatitis B virus infection, alpha-fetoprotein, Child-Pugh grade, Barcelona Clinic Liver Cancer stage, and platelet count;

[0067] (2) Imaging indicators, including: tumor number, tumor length (≤3 cm or >3 cm), tumor location (special location refers to tumor adjacent to the diaphragm, gallbladder, gastrointestinal tract, bile duct or located in the first hepatic hilar region, under the liver capsule, etc.), and liver cirrhosis.

[0068] Furthermore, the description of MRI examination, lesion selection and annotation in this embodiment is as follows:

[0069] All patients underwent abdominal dynamic enhanced MRI examination within 1 week before treatment. The German Siemens Skyra 3.0T MRI scanning system and the American CE Singa HDx 3.0 T MRI scanning system were used, both using 16-channel phased array coils. The scanning sequences included: transverse axial fat suppression T2-weighted imaging sequence, respiratory triggered transverse axial single-shot echo-planar diffusion imaging sequence, breath-hold transverse axial in-phase and anti-phase dual echo sequence, breath-hold transverse axial rapid three-dimensional spoiled gradient echo T2-weighted imaging sequence. Subsequently, contrast agent (gadopentetate dimeglumine) was injected through the cubital vein by a high-pressure injector, and scanning was performed in the arterial phase (18-22s), portal venous phase (60-75s), and delayed phase (90-180s).

[0070] The delineation of the region of interest was completed by two radiologists using IKSNAP software. On the axial T-weighted imaging sequence, diffusion-weighted imaging sequence, arterial phase, portal venous phase, and delayed phase images of the MRI, the whole layer was manually delineated and annotated along the lesion contour with an indentation of about 0.1 mm. When delineating, the complete lesion (including necrosis, hemorrhage, and cystic changes in the tumor) needs to be included, while avoiding surrounding blood vessels and satellite nodules, and another chief radiologist will review it.

[0071] This embodiment uses the United Imaging Intelligent Research Platform System V1.0 (developed by Beijing United Imaging Intelligent Imaging Technology Research Institute) for database management, incorporates patient data into the system, and then performs three-dimensional volume feature extraction, feature selection, and machine learning model construction. The training set performs feature selection and machine learning model construction, and the test set evaluates the performance of the constructed model.

[0072] Preferably, performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image comprises:

[0073] Performing filtering on the adjusted image to obtain a filtered image;

[0074] Construct an image enhancement window, and use the image enhancement window to traverse the filtered image for denoising, to obtain a denoised image;

[0075] The denoised image is subjected to histogram equalization processing to obtain an enhanced image.

[0076] Preferably, filtering the adjusted image to obtain a filtered image comprises:

[0077] Performing wavelet decomposition on the adjusted image at multiple scales to obtain corresponding wavelet coefficients;

[0078] constructing a filtering threshold according to the size and decomposition scale of the adjusted image;

[0079] constructing an image filtering model using the filtering threshold;

[0080] The adjusted image is filtered using the image filtering model to obtain the filtered image.

[0081] Preferably, the calculation formula of the filtering threshold is: Among them, λ j represents the filtering threshold at the jth decomposition scale, σ j represents the standard deviation of the noise at the jth decomposition scale, N j represents the signal length at the jth decomposition scale, and j represents the decomposition scale.

[0082] Preferably, the calculation formula of the image filtering model is:

[0083] Using the formula:

[0084]

[0085] Wherein, sign is the sign function, a is the first preset coefficient, b is the second preset coefficient, ω j,k represents the kth wavelet coefficient at the jth decomposition scale, Represents the wavelet coefficients after filtering.

[0086] Preferably, the data enhancement includes rotation, scaling, flipping and random cropping.

[0087] Specifically, the denoising of this embodiment is an important step to improve the quality of MRI images. First, the original MRI image is decomposed by wavelet transform to obtain wavelet coefficients of different scales. Then, according to the standard deviation of the noise and the characteristics of the signal, a suitable threshold is set, and the wavelet coefficients are filtered using the threshold to remove high-frequency noise. Finally, the denoised coefficients are reconstructed into a clear image through inverse wavelet transform. This process can effectively reduce the random noise in the image and improve the accuracy of subsequent processing.

[0088] Furthermore, normalization aims to eliminate the brightness and contrast differences between different MRI images. The specific method is to scale the pixel values ​​of each image to a uniform range (usually 0 to 1) to facilitate subsequent feature extraction and model training. Normalization can be achieved by subtracting the minimum value of the image from each pixel value and then dividing it by the difference between the maximum and minimum values ​​of the image. This process ensures that different images are compared at the same scale, avoiding the degradation of model performance due to brightness differences.

[0089] Furthermore, the resizing of this embodiment is to ensure that all input images have the same size for batch processing and model input. Typically, a standard size (e.g., 256×256 or 512×512 pixels) is selected and the original image is scaled using an interpolation method (e.g., bilinear interpolation or cubic interpolation). In this way, all images are resized to the same size, ensuring data consistency and facilitating subsequent feature extraction and model training.

[0090] Specifically, the image enhancement and data enhancement of this embodiment are important means to improve the generalization ability of the model. Image enhancement includes performing histogram equalization, contrast enhancement and other processing on the adjusted image to improve the visualization effect and feature recognizability of the image. Data enhancement generates diversified training samples by rotating, scaling, flipping, and randomly cropping the image. These enhancement techniques not only increase the amount of training data, but also improve the robustness of the model to different transformations, thereby improving the performance of the model in practical applications.

[0091] Optionally, before extracting image features, this embodiment also uses Z-score normalization and B-spline interpolation resampling algorithm technology to preprocess images of each sequence image of each patient to standardize the image data. Subsequently, 2264 quantitative omics features in the three-dimensional volume of interest lesions of each patient T, weighted imaging sequence, diffusion-weighted imaging sequence, arterial phase, portal venous phase and delayed phase are extracted. The extracted imaging omics features are preprocessed using the Z-score normalization method. The repeatability of feature extraction is evaluated using the intraclass correlation coemicient (ICC), and ICC>0.75 is considered to be a relatively stable imaging omics feature. Variance threshold method, selectK-best and LASSO regression screening features are then used in turn, and the imaging omics feature subset with the most predictive value is finally screened. Among them, selectK-best method selects features with P value <0.05 and removes other features.

[0092] Optionally, the data preprocessing step of clinical characteristics in this embodiment includes:

[0093] 1. This embodiment preprocesses the collected clinical feature data to ensure the integrity and consistency of the data. The preprocessing steps include missing value filling and unique hot encoding. Missing value filling can be done by the mean filling or median filling method for numerical features (such as age, liver function indicators, etc.), and for categorical features (such as gender, tumor differentiation, etc.), the mode is used for filling. Next, the categorical features are uniquely encoded, and each category is converted into a binary feature for subsequent model input. This process of this embodiment ensures that all features are compared on the same scale, avoiding the degradation of model performance due to data inconsistency.

[0094] 2. After completing the data preprocessing, the present embodiment then performs feature selection to extract the most influential clinical features for predicting the recurrence of hepatocellular carcinoma. A variety of feature selection methods are used, such as variance threshold method, chi-square test, LASSO regression, etc. Through these methods, the present embodiment can evaluate the correlation between each feature and the risk of recurrence and screen out important features. For example, using LASSO regression can automatically select features through regularization technology, reduce the impact of redundant features, and thus improve the predictive performance of the model. Ultimately, the selected features will be used for subsequent feature extraction and model training.

[0095] 3. The feature extraction stage of this embodiment processes the preprocessed clinical feature data through a fully connected neural network (FCNN). First, the preprocessed feature data is input into the FCNN, and after nonlinear transformations of multiple layers of fully connected layers, potential feature representations are gradually extracted. The output of each layer is processed by an activation function (such as ReLU) to introduce nonlinear factors and enhance the expression ability of the model. Finally, after the last fully connected layer, a clinical feature vector of fixed dimension is obtained, which will be fused with the image features to provide a basis for subsequent recurrence risk prediction.

[0096] Preferably, the clinical characteristics include: preoperative alpha-fetoprotein and platelet count.

[0097] Specifically, this example used univariate and multivariate logistic regression to analyze the relationship between clinical indicators and early postoperative recurrence. Preoperative alpha-fetoprotein>20μg / L, platelet count>140×10 9 / L and the special location of the tumor are factors affecting the early recurrence of patients with hepatocellular carcinoma after radiofrequency ablation. Alpha-fetoprotein is a persistent marker for the diagnosis and prediction of prognosis of patients with hepatocellular carcinoma. Up to 70% of patients with primary hepatocellular carcinoma have abnormally elevated serum alpha-fetoprotein. Previous studies have shown that a decrease in preoperative platelet count is a risk factor for recurrence in patients with hepatocellular carcinoma after surgical resection. In addition, in this embodiment, when the tumor is adjacent to the diaphragm, gallbladder, or located in the first hepatic portal area, under the liver capsule, etc., it is more likely to recur early. This may be because when treating the above-mentioned special sites, in order to avoid the heat sink effect during the operation to damage adjacent tissues, organs or important anatomical structures, ablation is more cautious, resulting in the ablation effect not being able to fully achieve the expected effect, thereby increasing the potential risk of recurrence.

[0098] Optionally, in the feature fusion stage of this embodiment, the image features and clinical features are merged through a splicing operation. Specifically, the image feature vector and the clinical feature vector are spliced ​​in the feature dimension to form a new joint feature vector. In order to further enhance the expressive power of the features, a weighted splicing method can be used to assign different weights to each feature according to its importance in the prediction. In addition, a Transformer module can be introduced to process the spliced ​​features, and the complex relationship between the features can be captured through the self-attention mechanism, thereby generating more representative fusion features.

[0099] Specifically, the initial multimodal deep learning model of this embodiment is composed of multiple fully connected layers, which is responsible for mapping the fused feature vector to the prediction result of recurrence risk. First, the fused feature vector is input to the first fully connected layer, which is usually set to 128 neurons, and the ReLU activation function is used to introduce nonlinearity. Then, after the second fully connected layer, the dimension is further reduced to 64 neurons, and the ReLU activation function is still used. Finally, the output layer uses the Softmax activation function to map the output to two categories (recurrence and non-recurrence), and outputs the probability value of each category.

[0100] Figure 2 A schematic diagram of the system structure in an embodiment of the present invention is shown in FIG. Figure 2 As shown, the present invention also provides a hepatocellular carcinoma recurrence prediction system combining MRI images and clinical characteristics, comprising:

[0101] A data collection module, used to collect multimodal sample MRI image data and preset clinical feature data of patients with hepatocellular carcinoma;

[0102] An image data processing module, used for performing denoising, normalization and size adjustment on the multimodal sample MRI image data to obtain an adjusted image, and performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image;

[0103] A clinical data processing module, used for performing missing value filling and one-hot encoding on the clinical characteristic data to obtain pre-processed characteristic data;

[0104] A feature extraction module, used to extract features from the enhanced image and the preprocessed feature data respectively to obtain image features and clinical features;

[0105] A feature fusion module, used for fusing the image feature and the clinical feature to obtain a fusion feature;

[0106] A training module, used for inputting the fusion features into an initial multimodal deep learning model for training to obtain a hepatocellular carcinoma recurrence prediction model;

[0107] The prediction module is used to input the MRI image and clinical characteristics to be tested into the hepatocellular carcinoma recurrence prediction model to obtain the prediction result.

[0108] The beneficial effects of the present invention are as follows:

[0109] (1) The present invention integrates MRI image features and clinical features and makes full use of multiple information sources to more comprehensively reflect the patient's pathological status, thereby improving the accuracy of recurrence prediction. The use of data enhancement technology (such as image rotation, scaling, flipping, etc.) and multimodal data fusion can effectively reduce the risk of overfitting of the model and make its performance on different patient data more stable.

[0110] (2) The method of the present invention can provide each patient with a personalized recurrence risk assessment, helping doctors to develop more reasonable follow-up and treatment plans, thereby improving the patient's survival rate and quality of life. Through comprehensive analysis of multimodal MRI images (such as T1, T2 weighted imaging and diffusion weighted imaging) and clinical characteristics (such as age, gender, liver function indicators, etc.), it is possible to capture richer feature information and improve the prediction ability of the model.

[0111] (3) The present invention reduces the workload of manual feature selection and improves the efficiency and accuracy of the prediction process through automatic feature extraction and fusion of deep learning models.

[0112] (4) The prediction model of the present invention can be used as a clinical decision support tool to help doctors identify high-risk patients in the early postoperative period and take timely intervention measures to reduce the recurrence rate. The implementation of this method provides new ideas and methods for the study of hepatocellular carcinoma, promotes the research progress of combining imaging omics with clinical data, and has important academic value.

[0113] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0114] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features, characterized in that: include: Collect multimodal sample MRI image data and preset clinical feature data of patients with hepatocellular carcinoma; Denoising, normalizing and resizing the multimodal sample MRI image data to obtain an adjusted image, and performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image; Performing missing value filling and one-hot encoding on the clinical characteristic data to obtain preprocessed characteristic data; Performing feature extraction on the enhanced image and the preprocessed feature data respectively to obtain image features and clinical features; fusing the image features and the clinical features to obtain fused features; Inputting the fusion features into an initial multimodal deep learning model for training to obtain a hepatocellular carcinoma recurrence prediction model; The MRI images and clinical features to be tested are input into the hepatocellular carcinoma recurrence prediction model to obtain the prediction results.

2. The method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features according to claim 1, characterized in that: The multimodal sample MRI image data includes: T1-weighted imaging data, T2-weighted imaging data and diffusion-weighted imaging data.

3. The method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features according to claim 1, characterized in that: The clinical characteristic data include age, gender, liver function index, tumor size, degree of differentiation and AFP level.

4. The method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features according to claim 1, characterized in that: Performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image includes: Performing filtering on the adjusted image to obtain a filtered image; Construct an image enhancement window, and use the image enhancement window to traverse the filtered image for denoising, to obtain a denoised image; The denoised image is subjected to histogram equalization processing to obtain an enhanced image.

5. The method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features according to claim 4, characterized in that: Performing filtering on the adjusted image to obtain a filtered image includes: Performing wavelet decomposition on the adjusted image at multiple scales to obtain corresponding wavelet coefficients; constructing a filtering threshold according to the size and decomposition scale of the adjusted image; constructing an image filtering model using the filtering threshold; The adjusted image is filtered using the image filtering model to obtain the filtered image.

6. The method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features according to claim 4, characterized in that: The calculation formula of the filtering threshold is: Among them, λ j represents the filtering threshold at the jth decomposition scale, σ j represents the standard deviation of the noise at the jth decomposition scale, N j represents the signal length at the jth decomposition scale, and j represents the decomposition scale.

7. The method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features according to claim 6, characterized in that: The calculation formula of the image filtering model is: Using the formula: Wherein, sign is the sign function, a is the first preset coefficient, b is the second preset coefficient, ω j,k represents the kth wavelet coefficient at the jth decomposition scale, Represents the wavelet coefficients after filtering.

8. The method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features according to claim 1, characterized in that: The data augmentation includes rotation, scaling, flipping, and random cropping.

9. The method for predicting the recurrence of hepatocellular carcinoma by combining MRI images and clinical features according to claim 1, characterized in that: The clinical characteristics include: preoperative alpha-fetoprotein and platelet count.

10. A hepatocellular carcinoma recurrence prediction system combining MRI images and clinical features, characterized in that: include: A data collection module, used to collect multimodal sample MRI image data and preset clinical feature data of patients with hepatocellular carcinoma; An image data processing module, used for performing denoising, normalization and size adjustment on the multimodal sample MRI image data to obtain an adjusted image, and performing image enhancement and data enhancement on the adjusted image to obtain an enhanced image; A clinical data processing module, used for performing missing value filling and one-hot encoding on the clinical characteristic data to obtain pre-processed characteristic data; A feature extraction module, used to extract features from the enhanced image and the preprocessed feature data respectively to obtain image features and clinical features; A feature fusion module, used for fusing the image feature and the clinical feature to obtain a fusion feature; A training module, used for inputting the fusion features into an initial multimodal deep learning model for training to obtain a hepatocellular carcinoma recurrence prediction model; The prediction module is used to input the MRI image and clinical characteristics to be tested into the hepatocellular carcinoma recurrence prediction model to obtain the prediction result.

Images