Hepatocellular Carcinoma Grading via Combined Computer Vision and Radiomics Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for pathological grading of hepatocellular carcinoma, such as tissue biopsy, are invasive and inaccurate, and radiomics features are prone to noise and low-order image limitations, necessitating a non-invasive and more precise evaluation tool.
Innovation Solution
A method combining computer vision features, including LoG, LBP, HOG, and Haar-like features, with radiomics features like morphological, grey scale, texture, and wavelet features, to create a predictive model for hepatocellular carcinoma grading through preprocessing, feature extraction, screening, and multivariable logistic regression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If tissue biopsy is used for pathological grading, then grading can be achieved, but it is invasive and has accuracy controversies
Solution Approach 1:
The patent replaces the mechanical invasive biopsy procedure with a non-invasive image-based radiomics analysis system. CT images are processed through feature extraction (morphological, texture, wavelet, grey-scale features) and fed into machine learning models (random forest, support vector machine, neural network) to predict pathological grading, eliminating the need for physical tissue sampling while achieving comparable or superior accuracy.
Solution Approach 2:
The patent creates a virtual copy of the pathological grading process by training machine learning models on radiomics features extracted from CT images. The models learn the mapping between image features and pathological grades, enabling non-invasive prediction that replicates the diagnostic value of actual biopsy without the associated risks.
2Ease of manufacture
If manual radiomics features are extracted using mathematical formulas, then feature extraction is simple, but the features are easily affected by noise and low-order image features
Solution Approach 1:
The patent combines multiple types of radiomics features (morphological, texture, wavelet, grey-scale) into a composite feature set. This multi-component approach compensates for the weaknesses of individual feature types by leveraging their complementary strengths, creating a more robust and noise-resistant feature representation for pathological grading prediction.
Solution Approach 2:
The patent transforms the feature extraction process by applying multiple mathematical transformations (wavelet transforms, texture analysis operators, morphological operations) to the same image data. This generates diverse feature representations that capture different aspects of the tumor characteristics, making the overall system less sensitive to noise in any single feature domain.
3Reliability
If computer vision features are used instead of manual features, then rotation invariance and noise insensitivity are improved, but the system complexity increases
Solution Approach 1:
The patent divides the complex feature extraction process into distinct modular components: morphological feature extraction, texture feature extraction, wavelet feature extraction, and grey-scale feature extraction. Each module handles a specific aspect of feature extraction with dedicated algorithms, making the overall complex system more manageable and maintainable while preserving the robustness benefits of computer vision features.
Data Source
AI summary
The present disclosure discloses a method and equipment for classifying hepatocellular carcinoma images by combining computer vision features and radiomics features, wherein the method comprising: 1) collecting eligible clinical images of patients and preprocessing the collected images; 2) extracting computer vision features from a segmented image of a hepatic tumor region; 3) extracting the manual radiomics features from the segmented image of the hepatic tumor region; 4) by combining the computer vision features and the radiomics features, screening by univariate filtering and then by LASSO regression; 5) using the features resulted from screening and clinical features together for modeling by a multivariable logistic regression model, and using the Akaike information criterion (AIC) to search backward and select clinical features suitable for the best model, so as to implement the prediction of hepatocellular carcinoma pathological grading.