Generative Model for 3D Medical Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging technologies face challenges in generating accurate 3D images of human body parts, particularly in comparing healthy and affected conditions due to differences in imaging apparatuses, artifacts, and physiological variations among individuals, which hinders effective diagnosis and prognosis.
Innovation Solution
A computer-implemented method using a generative network trained on a library of 3D scans to generate a complete 3D image of a human body part by optimizing a latent variable to minimize the distance between affected and non-affected image subsets, allowing for comparison of affected areas with healthy or simulated conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If medical images are taken using different imaging apparatuses, then more imaging options and flexibility are available, but image comparison becomes impossible due to different formats and resolutions
Solution Approach 1:
The patent introduces a generative model as an intermediary that translates images from different imaging apparatuses into a unified reference format. This mediator enables comparison by converting diverse input formats (different resolutions, formats, and quality levels) into a common output format, thereby resolving the incompatibility issue while preserving imaging versatility
Solution Approach 2:
The generative model dynamically adjusts image parameters (resolution, format, quality) to transform input images into a standardized reference format. By changing these parameters adaptively based on the input characteristics, the system enables accurate comparison across different imaging apparatuses without losing the flexibility to use various imaging options
2Loss of time
If medical images are taken at different time points, then longitudinal evolution can be observed, but accurate comparison is prevented by pathology development and image quality variations
Solution Approach 1:
The system performs preliminary normalization and quality enhancement on historical images before comparison. By pre-processing images to a unified reference format and applying quality enhancement techniques, the system prepares images for accurate temporal comparison, enabling longitudinal observation while mitigating quality variations that would otherwise prevent precise measurement
Solution Approach 2:
The generative model serves as a temporal intermediary that harmonizes images taken at different time points. It converts historical images into the reference format and enhances their quality, creating a consistent basis for comparison that preserves temporal evolution information while eliminating quality discrepancies
3Loss of information
If complete 3D images are generated including affected areas, then comprehensive diagnosis is possible, but artifacts and pathology alter the image quality and prevent accurate observation
Solution Approach 1:
The system extracts and isolates affected areas from the complete 3D image while preserving the rest of the anatomy. By separating the affected regions and processing them through the generative model, the system maintains comprehensive diagnostic information while enhancing the quality and observability of the affected areas, effectively removing the detrimental impact of artifacts and pathology on image quality
Data Source
AI summary
A method for generating a 3D image of a human body part including: train a generative network based on a library of human body part 3D scans of reference to obtain a generative model; make a 3D scan of a studied human body part; define a subset of the studied 3D scan by excluding the content of an area; optimize a latent variable for minimizing a distance between the defined subset and an image of a subset generated by the generative model from the latent variable; generate a complete 3D image of the studied human body part with the generative model using the optimized latent variable.