Image Fusion Using ML-Determined Scanning Geometry
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging technologies face challenges in generating high-quality fusion images efficiently, particularly in cancer management, where lower resolution images after treatment need to be fused with higher resolution reference images, often requiring external tracking systems and radiation exposure.
Innovation Solution
A medical imaging device that uses a machine learning algorithm to determine specific image scanning geometries from lower resolution images, allowing for the generation of fusion images without external tracking systems, by encoding image scanning geometry in the images and using feature-based registration, thereby reducing radiation exposure and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If external tracking systems are used to determine image scanning geometry, then registration accuracy is improved, but device complexity and cost increase
Solution Approach 1:
The patent applies self-service by enabling the lower resolution image scanning unit to encode its own image scanning geometry information directly within the acquired images. The image processing unit then extracts this geometry information autonomously without requiring external tracking systems, allowing the system to determine its own spatial parameters through self-contained image data analysis
Solution Approach 2:
The patent introduces image data as an intermediary carrier that contains encoded scanning geometry information. Instead of using direct external tracking hardware, the system uses the image data itself as a mediator to convey spatial and geometric information from the scanning unit to the processing unit, eliminating the need for separate tracking devices
2Measurement precision
If higher resolution reference images are acquired frequently, then image quality is improved, but radiation exposure increases
Solution Approach 1:
The patent applies parameter changes by transforming the resolution parameter of reference images through super-resolution algorithms. The system takes multiple lower resolution images and processes them to generate higher resolution reference images, thereby changing the resolution parameter without requiring additional high-resolution acquisitions that would increase radiation exposure
Solution Approach 2:
The patent uses copying by creating enhanced versions of reference images through computational methods. Instead of acquiring new high-resolution images that would expose the patient to more radiation, the system copies and enhances existing lower resolution images using super-resolution algorithms to produce high-quality reference images for fusion
3Device complexity
If feature-based registration is used instead of external tracking, then device complexity is reduced, but measurement precision may worsen
Solution Approach 1:
The patent applies dimensionality change by extracting and utilizing image scanning geometry information that exists within the image data itself. Instead of relying on external spatial tracking, the system analyzes geometric parameters embedded in the image acquisition process, adding a new dimension of information extraction from the image data to achieve accurate registration
Data Source
AI summary
Described are various embodiments of systems, devices, methods, and computer readable mediums related to image fusion of lower resolution (LR) images and higher resolution (HR) reference images of an object in a body. In an example embodiment of a method, the method comprises determining specific image scanning geometries used for acquiring LR images based on an ML algorithm that uses at least one feature in at least one region of interest in the respective LR images as input. The method further comprises generating image scanning geometry matching oblique HR images based on HR reference images and the determined specific image scanning geometries used for acquiring the respective LR images and registering the oblique HR images with the LR images to generate registered HR images. Current feature information is extracted from the LR images and mapped on corresponding feature information in the registered HR images to generate fusion images.