Tissue Dissection Annotation Transfer via Bidirectional Registration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for transferring annotations for tissue dissection from stained reference slides to unstained extraction slides are insufficiently robust, especially when the biological material has varying shapes and positions across slices.
Innovation Solution
A computer-implemented method and system that determines regions for tissue dissection by obtaining annotations from multiple reference images, registering them with intermediate images, and combining these annotations using registration parameters to generate more robust annotations for tissue dissection, accounting for variations in shape and position across a series of pathology slides.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If annotation transfer is performed using a single reference image, then the process is simple and quick, but the accuracy and robustness deteriorate when tissue shapes and positions vary across slices
Solution Approach 1:
The annotation transfer process is segmented into multiple independent steps: obtaining annotations from multiple reference images, registering each reference image with the target image using registration parameters, propagating annotations from each reference image, and combining the propagated annotations. This segmentation allows each step to be optimized independently while maintaining overall robustness.
Solution Approach 2:
Multiple annotations from different reference images are merged into a single combined annotation for the target image. The combination process integrates information from multiple sources, improving the reliability and robustness of the final annotation by accounting for variations in tissue shapes and positions across different reference images.
2Measurement precision
If multiple reference images are used for annotation transfer, then the accuracy and robustness improve, but the processing time and computational resources increase
Solution Approach 1:
Registration parameters between reference images and the target image are determined in advance before annotation propagation. This preliminary registration step allows for efficient annotation transfer by pre-establishing the spatial relationships, reducing the computational burden during the actual annotation propagation and combination phases.
Solution Approach 2:
Annotations from reference images are copied and propagated to the target image using determined registration parameters. This copying approach allows rapid transfer of annotation information once registration is established, reducing processing time while maintaining accuracy through the use of multiple reference images.
3Reliability
If registration parameters are determined for each intermediate image with multiple reference images, then the annotation robustness improves, but the computational complexity increases
Solution Approach 1:
Different registration parameters are determined for each reference image based on its specific characteristics and relationship with the target image. This local optimization approach ensures that each registration is tailored to the specific geometric and morphological features of that reference image, improving overall robustness while managing computational complexity through targeted processing.
Data Source
AI summary
Some embodiments are directed to a system and computer-implemented method are provided for determining one or more regions for tissue dissection in a series of pathology slides using a series of images which represent a digitized version of the series of pathology slides. Annotations are obtained for at least two reference images, with each annotation representing a region for tissue dissection in the respective reference image. Annotations are then generated for intermediate images between the reference images on the basis of bidirectional image registration and the subsequent propagation of both annotations to each intermediate image, which annotations are then combined to obtain a combined annotation for each intermediate image. The above measures are well suited to generate annotations for series of pathology slides which contain tissue slices of a tissue of interest having a complex 3D shape.