Validity Evaluation Device for Cancer Region Detection in MRI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cancer detection methods using deep learning models struggle to accurately validate cancer regions in MRI images, as they often fail to distinguish between similar local features and provide unreliable recognition results, especially when the disease manifests in various forms.
Innovation Solution
A method and apparatus that utilize parametric MRI and a cancer region detection model to generate guide information for user input, allowing for the comparison of detected cancer regions with pathology images to evaluate validity, thereby quantitatively assessing the coincidence between deep learning-based detections and actual cancer regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If deep learning models are used to detect cancer regions in MRI images, then automated detection capability is improved, but reliability of detection results deteriorates due to inability to distinguish similar local features
Solution Approach 1:
The patent introduces a validity evaluation module as an intermediary between the deep learning detection model and the final diagnosis. This module compares the detected cancer region with the original MRI image and calculates a validity score, acting as a mediator that verifies the reliability of automated detection results before they are finalized
Solution Approach 2:
The system implements a feedback mechanism where the validity evaluation results are fed back to adjust or verify the detection outcomes. The validity score calculated by comparing detected regions with original images provides feedback that can confirm or reject automated detection results, improving overall reliability
2Productivity
If deep learning models rely on local feature extraction using CNN and max pooling, then processing efficiency is improved, but recognition accuracy deteriorates for images with different content but similar local information
Solution Approach 1:
The patent moves from purely local feature analysis to a multi-dimensional approach by incorporating global context analysis. The validity evaluation module examines the detected cancer region in the context of the entire MRI image, adding a global dimension to the detection process that complements the local feature extraction
3Loss of information
If parametric MRI is used to represent body changes and disease characteristics, then information richness is improved, but difficulty in formalizing correlation with actual disease deteriorates
Solution Approach 1:
The system transforms the complex parametric MRI data into a validity score parameter that quantifies the correlation between detected regions and actual disease. By changing the parameter representation from multiple MRI parameters to a single validity metric, the system simplifies the formalization of disease correlation
Data Source
AI summary
An apparatus for evaluating validity of detection of a cancer region may be provided. The apparatus comprises a parametric magnetic resonance imaging (MRI) provider configured to provide at least one MRI constructed based on different parameters, a first cancer region input unit configured to receive a first cancer region based on the at least one parametric MRI, a cancer region processor including a cancer region detection model for receiving the at least one parametric MRI as input and outputting cancer region information and configured to generate and provide guide information corresponding to an image to be analyzed through the cancer region detection model, a second cancer region input unit configured to receive a second cancer region based on the guide information, and a validity evaluator configured to generate validity evaluation information of the second cancer region, by comparing the first cancer region with the second cancer region based on a pathology image obtained by mapping a region, in which cancer is present, of an extracted body portion.