CT, MRI, and PET Fusion for Organ-at-Risk Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing radiotherapy methods rely heavily on manual CT image delineation, leading to high workload, decision fatigue, errors, and low efficiency due to image repetition and complexity, particularly for small organs, with CT-based auto-segmentation systems exhibiting low precision and effectiveness.
Innovation Solution
An image processing system that merges CT, MRI, and PET images using a machine learning algorithm to form a fused image, employing a residual network with channel attention and 3D convolution, and utilizes multimodal fusion and segmentation models for accurate and efficient organ-at-risk segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual CT image delineation is used, then organ segmentation can be performed, but workload is high and efficiency is low due to image repetition and decision fatigue
Solution Approach 1:
The system enables automated self-service organ segmentation by training a machine learning model to perform delineation tasks autonomously. The model processes CT images and automatically generates segmentation contours without requiring manual radiologist intervention for each case, thereby dramatically improving productivity and reducing time loss.
Solution Approach 2:
The patent replaces the mechanical manual delineation process with an automated machine learning-based system. The mechanical action of manual contour drawing is substituted by computational algorithms that automatically identify and segment organs, eliminating decision fatigue and repetitive manual labor while maintaining or improving segmentation quality.
2Productivity
If CT-based auto-segmentation is used, then efficiency is improved, but precision is low due to artifacts and image complexity particularly for soft tissue organs
Solution Approach 1:
The system changes key parameters of the segmentation process by training the machine learning model on extensively annotated CT images to learn optimal segmentation parameters. The model adapts to various imaging conditions, artifacts, and anatomical variations, dynamically adjusting segmentation thresholds and parameters to maintain high precision across different cases while preserving automated efficiency.
Solution Approach 2:
The patent incorporates feedback mechanisms where the machine learning model continuously learns from annotated segmentation results. The feedback from radiologist corrections and validation cases is used to retrain and refine the model, progressively improving segmentation precision for soft tissue organs and reducing errors caused by image artifacts and complexity.
3Ease of operation
If automated segmentation is used, then workload is reduced, but errors increase due to fatigue and lack of experience in manual delineation
Solution Approach 1:
The automated segmentation system performs self-service by autonomously processing images and generating segmentation results without human intervention. This eliminates operational errors caused by fatigue and inexperience, as the machine learning model consistently applies learned segmentation patterns. The system maintains high reliability through continuous training on diverse annotated data and incorporates validation mechanisms to ensure consistent, error-free operation.
Data Source
AI summary
An image processing system for organ-at-risk segmentation includes a database, a data input module, a machine learning module, and an image processing module. The database stores medical images of organs. The machine learning module stores a machine learning algorithm that algorithmically generates a merging model, an extraction model, and a segmentation model for each organ based on the medical images of the organs. The data input module inputs data under test including at least two input medical images of different types. The image processing module merges the at least two input medical images to form a fused image based on the merging model, extracts at least one key feature of the fused image based on the extraction model, identifies the key feature extracted based on the segmentation model, and delineates contours of the organ corresponding to the key feature in the fused image based on the identification result.