Liver Lesion Segmentation via Multi-Phase Attention Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing liver lesion segmentation methods fail to effectively handle inconsistencies in multi-phase images due to registration errors from respiration-induced motion, differences in scanner settings, unavailability of segmentation ground-truth, and invisible regions of interest.
Innovation Solution
A deep learning-based radiomics method that preprocesses and augments arterial and portal phase images using multi-channel inputs, co-registration, interpolation, and adaptive histogram equalization to improve segmentation accuracy, particularly for liver cancer diagnosis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multi-phase images are used for liver lesion segmentation, then more information is available for analysis, but registration errors due to respiration-induced motion and scanner setting differences worsen segmentation accuracy
Solution Approach 1:
The patent introduces an attention gate mechanism as an intermediary component that selectively filters and weights features from different phases. This attention gate acts as a mediator between the multi-phase inputs and the segmentation output, allowing the model to automatically identify and emphasize reliable features while suppressing registration errors and inconsistencies from misaligned phases.
Solution Approach 2:
The patent dynamically adjusts the weighting parameters of different phase images through the attention mechanism. By changing the importance weights assigned to each phase based on their reliability and alignment quality, the model adapts to varying registration errors and scanner settings, optimizing segmentation accuracy despite parameter variations across phases.
2Reliability
If deep learning models are trained on inconsistent multi-phase images, then robustness to real-world variations improves, but training complexity and data preprocessing requirements increase
Solution Approach 1:
The patent performs preliminary registration and normalization of multi-phase images before feeding them to the deep learning model. By pre-aligning the phases and standardizing their parameters, the model receives pre-processed input that reduces the burden during training, while still maintaining robustness to real-world variations through the attention mechanism that handles residual inconsistencies.
3Measurement precision
If manual segmentation ground-truth is available for all phases, then segmentation accuracy improves, but the time and resources required for annotation increase
Solution Approach 1:
The patent leverages the ground-truth segmentation from a single reference phase and copies/adapts it to other phases through the attention mechanism and feature fusion. Instead of requiring manual annotation for each phase, the model learns to transfer the ground-truth information across phases by attending to phase-specific features, significantly reducing annotation time while maintaining accuracy.
Data Source
AI summary
Methods and systems for biomedical image segmentation based on a combination of arterial and portal image information Methods and systems for biomedical image segmentation based on a combination of arterial and portal image information are described. The combination of arterial and portal image information is helpful in improving biomedical image segmentation when the different phases of images are not properly registered for example due to respiration-induced motion of a patient or one of the phases have missing manual reference. A preferred embodiment is the segmentation or prediction of liver cancer or hepatocellular carcinoma.