Dual-Model Image Processing for Subtle Lesion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer-aided diagnosis (CAD) systems struggle to detect lesions that are not clearly shown in medical images, particularly in non-contrast tomographic images, as they are developed under the assumption that lesions are visibly present.
Innovation Solution
An image processing apparatus using a first derivation model to determine the likelihood of a region of interest for each pixel and a second derivation model to derive a predictive value indicating the presence of a specific finding, both constructed through machine learning on neural networks, with restrictions on the sum and threshold of pixel certainty to enhance detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If CAD systems are developed based on the assumption that lesions are clearly visible in medical images, then the detection accuracy for clearly visible lesions is improved, but the ability to detect lesions that are not clearly shown deteriorates
Solution Approach 1:
The patent segments the detection process into two distinct stages: first detecting clearly visible lesions using traditional CAD methods, and second detecting hardly shown lesions by analyzing indirect findings. This segmentation allows each detection stage to be optimized for its specific target, resolving the contradiction between detecting obvious and subtle lesions.
Solution Approach 2:
The patent introduces indirect findings as an intermediary element for detecting hardly shown lesions. Instead of directly detecting the lesion itself, the system detects changes in peripheral tissue (indirect findings) that serve as clues to the presence of the lesion, thereby enabling detection of lesions that are not clearly visible in the original image.
2Device complexity
If traditional CAD methods are used that focus on direct lesion detection, then the system complexity remains low, but the diagnostic capability for subtle lesions deteriorates
Solution Approach 1:
The patent divides the diagnostic system into two modules: a traditional CAD module for direct lesion detection and a new module for indirect finding-based detection. This segmentation allows the system to incorporate advanced diagnostic capabilities while maintaining the simplicity and reliability of traditional methods for cases where they are sufficient.
Solution Approach 2:
The patent creates a universal diagnostic system that can handle both clearly visible and hardly shown lesions through its dual-module architecture. The system automatically selects or combines the appropriate detection method based on the characteristics of the input image, providing multi-functional diagnostic capability without requiring separate specialized systems.
3Speed
If the CAD system analyzes only the lesion region directly, then the processing speed is high, but the accuracy for indirectly indicated lesions deteriorates
Solution Approach 1:
The patent performs preliminary analysis of indirect findings in the peripheral tissue before finalizing the lesion detection. By pre-identifying changes such as atrophy, swelling, stenosis, or calcification in surrounding tissues, the system prepares clues that guide the subsequent lesion detection process, improving accuracy for indirectly indicated lesions without significantly impacting processing speed.
Data Source
AI summary
A processor, derives a likelihood of a region of interest for each pixel of an input image via a first derivation model, and derives a predictive value representing a possibility that a specific finding is included in the input image from the input image and the likelihood of the region of interest via a second derivation model, in which the region of interest is a region serving as a basis for obtaining the predictive value.