Longitudinal Medical Image Comparison Using Finding-Based Change Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine-learning models for medical imaging are limited in their ability to effectively compare longitudinal studies, particularly in identifying subtle changes in medical findings over time, leading to inaccurate readings and increased workload for clinicians.
Innovation Solution
A computer-implemented method that utilizes a machine learning algorithm to generate a change map by selecting image pairs based on medical findings, using a language model to extract relevant information from medical reports, and applying image registration to ensure accurate comparison of medical images from different sessions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If machine-learning models are used to compare longitudinal medical images, then processing speed is improved, but the ability to identify subtle changes and diagnostic accuracy deteriorates
Solution Approach 1:
The patent segments the comparison task by first identifying specific medical findings from reports, then selecting relevant image pairs based on those findings. This segmentation allows the system to focus computational resources on specific regions and temporal comparisons that are most likely to show clinically significant changes, improving both speed and accuracy.
Solution Approach 2:
The system performs preliminary actions by extracting medical findings from reports and using them to pre-select relevant image pairs before conducting the full comparison. This preliminary filtering reduces the search space and allows the machine-learning model to focus on the most promising comparisons, enhancing diagnostic accuracy while maintaining processing speed.
2Reliability
If all medical images from longitudinal studies are compared, then comprehensive assessment is improved, but time required for analysis increases
Solution Approach 1:
The patent extracts only the most relevant information by first pulling medical findings from reports and then using those findings to select specific image pairs for comparison. This extraction approach filters out unnecessary images and focuses analysis on the most clinically significant comparisons, maintaining comprehensive assessment while reducing analysis time.
Solution Approach 2:
The system applies local quality by tailoring the comparison process to specific medical findings and their spatial locations. Rather than uniformly processing all images, the system concentrates computational effort on image pairs and regions that are most relevant to the identified findings, improving efficiency without sacrificing comprehensive assessment.
3Ease of operation
If generic machine-learning models are used for image comparison, then ease of operation is improved, but adaptability to specific medical findings deteriorates
Solution Approach 1:
The patent implements universality by designing a system that can handle multiple types of medical findings and image sequences through a unified framework. The machine-learning model is trained to compare images based on extracted medical findings, making it adaptable to various conditions and imaging protocols while maintaining ease of operation through automated finding-based selection.
Solution Approach 2:
The system uses feedback from medical reports to dynamically adjust the image selection and comparison process. By continuously incorporating finding information from reports into the image pair selection, the system adapts to specific medical conditions while maintaining user-friendly operation through automated adjustment rather than manual reconfiguration.
Data Source
Figure 1
Figure 2
Figure 3~5
AI summary
The invention describes a computer-implemented method for generating a change map indicating at least one difference between a current image and a prior image. The method comprises - obtaining a plurality of prior medical image sequences (30) from a prior study (30, 31), and a prior report (31) from the prior study (30,31); - obtaining a plurality of current medical image sequences (20) from a current study; - determining at least one medical finding based on the prior report; - identifying, based on the determined medical finding, an image pair (14P, 14C), wherein the image pair (14P, 14C) comprises a prior image (14P) from one of the prior medical image sequences (30) and a current image (14C) from one of the current medical image sequences (20), and - inputting the image pair (14P, 14C), and optionally the prior report (31), to a machine learning algorithm (16) trained to generate a change map (180) indicating at least one difference (180X) between the current image (14C) and the prior image (14P) of the image pair (14P, 14C).