CV/LLM Compressed Image Representation for Rapid Medical Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Medical images from patient exams are large files that are time-consuming to access due to the size of the image files and constraints placed by the DICOM standard, making it difficult for clinicians to view prior images during real-time scanning.
Innovation Solution
An image comparison system that uses a computer vision-enabled large language model (CV/LLM) to generate compressed representations (CRs) of medical images, allowing for rapid text-based comparisons between current and prior images without the need to retrieve the entire image files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clinicians retrieve and compare full medical image files from prior exams, then accurate anatomical comparison is achieved, but the process becomes time-consuming due to large file sizes and DICOM standard constraints
Solution Approach 1:
The system extracts only the essential visual information from full medical images by generating compressed representations (CRs) that capture key anatomical features. This extraction allows comparison of critical diagnostic elements without retrieving entire large-scale image files, thus maintaining anatomical comparison accuracy while dramatically reducing access time.
Solution Approach 2:
Instead of working with original full-resolution medical images, the system creates compressed representation copies that preserve essential diagnostic information. These CR copies enable rapid comparison operations while the full images remain stored in the archive, balancing fidelity with speed.
2Loss of information
If the system stores and processes full medical image files for comparison, then complete image information is available, but system resource requirements and processing complexity increase
Solution Approach 1:
The system segments medical image information into two distinct components: compressed representations (CRs) for rapid comparison operations and full original images stored in the archive. This segmentation allows the comparison system to work with lightweight CR data structures while preserving access to complete image information when needed, reducing processing complexity without sacrificing information completeness.
Solution Approach 2:
The compressed representation acts as an intermediary between the full medical images and the comparison process. Rather than directly processing large image files, the system uses CRs as a intermediate layer that captures essential visual information in a compact format, reducing computational burden while maintaining diagnostic relevance.
Data Source
AI summary
The current disclosure provides systems and methods for an automated image comparison system that generates natural language descriptions of images. In an example, the image comparison system is configured to carry out a method that includes acquiring a current image of a patient during a current exam, generating, with a computer vision-enabled large language model (CV/LLM), a first compressed representation (CR) of the current image, obtaining a second CR of a similar image, the similar image similar to the current image and acquired in a prior exam, generating a text-based comparison of the current image and the similar image using the CV/LLM by entering the first CR and the second CR as input to the CV/LLM, and outputting the text-based comparison.