Anatomical Feature Transplantation for Diverse MRI Segmentation Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning-based models for whole knee joint tissue segmentation, particularly in MRI, face accuracy issues due to insufficient diversity in training data, leading to incorrect segmentations when encountering unseen image features like meniscus lesions.
Innovation Solution
A method and system for anatomical features transplantation, involving spatial matching and augmentation of medical imaging data by transplanting regions of interest from a source to a destination volume, converting single masks to multi-class masks, and using intersection masks to enhance data variety for training deep learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning models are trained on limited medical imaging data, then training time and computational resources are reduced, but segmentation accuracy decreases when encountering unseen features like meniscus lesions
Solution Approach 1:
The patent creates synthetic training data by copying and transplanting anatomical regions (e.g., meniscus with lesions) from source medical images to destination images. This copying process generates diverse training samples without requiring additional real patient scans, thereby improving model reliability on unseen features while avoiding the need to collect more actual medical data.
Solution Approach 2:
The system transforms training data by applying various parameter changes including spatial transformations (rotation, flipping, scaling), intensity adjustments, and anatomical feature transplantation. These parameter modifications create augmented training datasets that expose the model to varied representations of anatomical structures, improving generalization accuracy without increasing the fundamental data quantity.
2Reliability
If more diverse training data is collected from additional patients, then model generalization improves, but patient privacy requirements and data collection costs increase
Solution Approach 1:
Instead of collecting data from additional patients, the system copies existing anatomical features and transplants them into different imaging contexts. This approach achieves model generalization by creating synthetic diversity from existing data, eliminating the need for complex multi-center data collection protocols and patient recruitment processes.
Solution Approach 2:
The patent introduces an intermediary data augmentation process that sits between existing training data and the deep learning model. This intermediary system performs automated feature transplantation and image synthesis, serving as a mediator that generates diverse training samples without requiring direct access to additional patient data or complex data management infrastructure.
3Measurement precision
If manual segmentation labeling is performed on more images to increase training diversity, then segmentation precision improves, but time and labor costs increase significantly
Solution Approach 1:
The system copies manually labeled segmentation masks from source images and transplants them to corresponding regions in destination images. This copying approach propagates existing high-quality annotations to generate additional labeled training data without requiring manual re-labeling, thereby maintaining segmentation precision while avoiding time-consuming manual annotation work.
Solution Approach 2:
The data augmentation system performs self-service by automatically generating labeled training data through feature transplantation and mask copying. The process requires no human intervention beyond the initial labeling phase, as the system autonomously creates diverse labeled samples by reusing existing annotations across different imaging contexts, eliminating the need for continuous manual labeling efforts.
Data Source
AI summary
A method includes segmenting a region of interest in an anatomical region in both a source imaging data and a destination imaging data, wherein different regions of the region of interest are labeled with different segmentation masks. The method includes selecting a region from the different regions from the source imaging data and spatially matching the region to a corresponding region in the destination imaging data. The method includes determining a spatial intersection between the region and the corresponding region. The method includes utilizing an intersection mask to crop a first portion of the region from the source imaging data and utilizing the intersection mask to remove a second portion of the destination imaging data in the corresponding region. The method includes adding the first portion of the region from the source imaging data into the destination imaging data where the second portion was removed.