Deep Learning Model for Hepatocellular Carcinoma Recurrence Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive models for hepatocellular carcinoma (HCC) recurrence after curative surgery are limited by their reliance on extracting features from regions of interest in CT images, which can lead to missed information and human error.
Innovation Solution
A novel deep-learning model that analyzes entire pre-operative CT images, using liver segmentation and a two-branches neural network architecture, to predict HCC recurrence risk by integrating features from hepatic arterial and portal venous phases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If features are extracted from regions of interest in CT images using traditional methods, then the modeling process is simpler and faster, but information may be missed and human error increases
Solution Approach 1:
The patent segments the CT imaging process into multiple phases (arterial phase and portal venous phase) and processes each phase separately through dedicated neural network branches, then combines the results. This segmentation allows comprehensive analysis of the entire liver while maintaining manageable computational complexity through modular architecture.
Solution Approach 2:
The deep learning model is designed to perform multiple functions: it processes both arterial and portal venous phase images, extracts features from the entire liver region, and generates recurrence risk predictions. This multi-functional approach eliminates the need for separate manual region-of-interest marking while maintaining comprehensive analysis.
2Loss of information
If the entire CT image is analyzed instead of limited regions of interest, then more information is captured and human error is reduced, but computational complexity and processing time increase
Solution Approach 1:
The model divides the entire CT image analysis into two separate processing branches corresponding to arterial and portal venous phases. Each branch independently processes its phase data through neural networks, then the results are combined. This segmentation enables comprehensive whole-liver analysis while maintaining efficient processing through parallel computation.
Solution Approach 2:
The model performs preliminary processing of CT images by automatically segmenting the liver region and separating arterial/portal venous phases before feature extraction. This preliminary organization of data prepares the entire image for efficient batch processing by the neural network, reducing overall processing time compared to manual region-of-interest identification.
3Measurement precision
If manual region of interest labeling is used, then the process is more interpretable and controllable, but human error and subjectivity increase
Solution Approach 1:
The deep learning model performs automatic liver segmentation and tumor region identification without requiring manual annotation by radiologists. The neural network independently processes the CT images, extracts relevant features, and generates predictions, eliminating human subjectivity and labeling errors while maintaining high precision through learned feature representations.
Solution Approach 2:
The patent replaces the manual mechanical process of region-of-interest labeling with an automated computational system. The neural network automatically identifies and segments relevant regions within the CT images, substituting human manual operations with algorithmic processing that eliminates human error and subjectivity while maintaining interpretability through visualizable feature maps.
Data Source
AI summary
The present invention relates to methods for predicting the recurrence risk of hepatocellular carcinoma (HCC). Specifically, it proposes a deep learning model capable of integrating information from different phases of CT images and clinical data to predict the risk of HCC recurrence within 1 to 5 years after treatment. Experimental results demonstrate that these models outperform traditional prediction methods based on histological microvascular invasion (MVI) in predicting HCC recurrence risk.