Medical Image Segmentation for Irregularity Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical image analysis using computer-implemented solutions faces challenges in accuracy due to the need to differentiate between relevant and irrelevant information in large medical images, leading to increased training time and potential loss of information during compression.
Innovation Solution
The method involves identifying and segmenting the body region of interest within medical images without significant compression, using these image segments to train a machine learning model, thereby reducing irrelevant information and improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If the resolution of medical images is reduced to decrease training time and computational resources, then training efficiency is improved, but information loss occurs which affects detection accuracy
Solution Approach 1:
The patent divides large medical images into multiple smaller image segments or patches. This segmentation allows the training process to work with smaller units that require less computational resources and time, while collectively covering the entire image. The model learns to detect irregularities in these segments and then integrates the results to make decisions about the complete image, thus reducing training time without losing critical diagnostic information.
2Productivity
If the size of medical images is reduced to improve processing speed, then training efficiency increases, but detection accuracy decreases
Solution Approach 1:
The patent processes large medical images by dividing them into smaller segments, enabling faster processing of each segment while maintaining the option to process segments at different resolutions. This allows the system to achieve high processing speeds through efficient parallel processing of multiple small segments while preserving detection accuracy by selectively maintaining high resolution in segments containing irregularities.
Solution Approach 2:
The patent introduces a hierarchical processing approach where image segments are processed at multiple resolution levels. The system first processes segments at a lower resolution for rapid screening, and then selectively processes specific segments at higher resolution when irregularities are detected. This multi-dimensional processing strategy maintains high processing speed while preserving detection accuracy when needed.
3Device complexity
If feature vectors are used to represent medical images for comparison, then computational complexity is reduced, but information loss occurs affecting analysis accuracy
Solution Approach 1:
The patent extracts and processes individual image segments rather than converting the entire medical image into a single feature vector. This approach allows the system to maintain the detailed visual information present in the original image segments while reducing computational complexity by processing smaller, manageable units. The raw pixel data in each segment is preserved and fed directly into the neural network, avoiding the information loss that occurs during feature vector extraction.
Data Source
AI summary
A method for training a computer-implemented machine learning model for detecting irregularities in medical images, the method including: identifying at least one predetermined type of body region (14) depicted in a medical image (10), said body region (14) having a depicted irregularity (12); defining a plurality of image segments (20) each including at least part of the depicted body region (14), wherein a resolution of the image segments (20) is maintained or not reduced by more than 20% compared to the medical image (10); and using said image segments (20) to train a machine learning model to detect similar irregularities (12) in other medical images (10). Further, the invention relates to a use and to systems for training a computer-implemented machine learning model for detecting irregularities in medical images.