CBAM Deep Learning Model for Non-Contrast CT HCC Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current diagnostic methods for hepatocellular carcinoma (HCC) using computed tomography (CT) scans rely on contrast-enhanced images, which can cause adverse reactions and are not definitive for intermediate categories, leading to delayed diagnosis and treatment, while non-contrast CT scans have limited diagnostic performance.
Innovation Solution
A deep neural network-based Convolutional Block Attention Module (CBAM) model processes and classifies non-contrast CT images using attention mechanisms to focus on relevant features, improving diagnostic accuracy for HCC without the need for contrast agents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If contrast-enhanced CT scans are used for HCC diagnosis, then diagnostic performance is improved, but adverse reactions and radiation exposure increase
Solution Approach 1:
The patent extracts and removes the contrast agent component from the diagnostic process by developing a deep learning model that processes only non-contrast CT images. The model learns to identify HCC characteristics without requiring iodinated contrast media, thereby eliminating contrast-induced adverse reactions while maintaining diagnostic capability through automated feature extraction from raw imaging data
Solution Approach 2:
The patent replaces the physiological enhancement mechanism (contrast agent circulation and tissue perfusion) with an computational intelligence system. The deep learning model substitutes the biological/chemical contrast mechanism with algorithmic pattern recognition, using convolutional neural networks to detect tumor characteristics directly from non-contrast images, thereby eliminating the need for contrast agents and their associated harms
2Adaptability or versatility
If intermediate LI-RADS categories (LR-2, LR-3, LR-4) are used for diagnosis, then diagnostic coverage is improved, but diagnosis and treatment are delayed
Solution Approach 1:
The patent implements a feedback mechanism where the deep learning model provides probabilistic diagnostic confidence scores for each detected lesion. When the model's confidence exceeds a predetermined threshold, immediate definitive diagnosis is achieved without requiring follow-up imaging. This feedback loop transforms intermediate uncertainty into actionable definitive diagnoses, eliminating the time loss associated with repeated surveillance imaging while maintaining comprehensive diagnostic coverage
Solution Approach 2:
The patent changes the diagnostic parameter from categorical LI-RADS classification (which creates intermediate uncertainty) to continuous probabilistic confidence scoring. The model outputs a probability distribution indicating the likelihood of HCC, allowing clinicians to set confidence thresholds that convert uncertain intermediate cases into definitive diagnoses, thereby reducing diagnostic delay while maintaining adaptability across different clinical scenarios
3Object-affected harmful factors
If non-contrast CT scans are used for HCC diagnosis, then adverse reactions are reduced, but diagnostic performance deteriorates
Solution Approach 1:
The patent replaces the physical contrast enhancement mechanism with computational intelligence. The deep learning model compensates for the lack of contrast-induced tissue differentiation by learning complex patterns in raw attenuation values, texture features, and spatial relationships within the liver parenchyma. This substitution maintains diagnostic performance while eliminating contrast agent requirements
Solution Approach 2:
The patent creates a universal diagnostic system that processes non-contrast CT images to achieve multiple diagnostic functions simultaneously: tumor detection, characterization, and confidence assessment. The single deep learning model performs what traditionally required contrast-enhanced imaging, making the system universally applicable to patients who cannot receive contrast agents while maintaining diagnostic accuracy
Data Source
AI summary
Disclosed is a computer-implemented three-dimensional image classification system (CIS) for processing and/or analyzing non-contrast computed tomography (CT) medical imaging data. The CIS is a deep neural network containing multiple Convolutional Block Attention Module (CBAM) blocks, which contain convolutional layers for feature extraction followed by CBAMs. The CBAM applies channel attention to highlight more relevant features and spatial attention to focus on more important regions. Max pooling layers operably link adjacent pairs of CBAM blocks. The output of the final CBAM block is passed to two terminal fully connected layers to generate a diagnosis. This classification system can be used to perform efficient diagnosis of hepatocellular carcinoma using solely non-contrast CT images, with diagnostic performance comparable to that of a radiologist using the current LIRADS system.