Methods for Reconnecting Fragmented Longitudinal Objects in 3D Micro-CT Data: With Application to Small-Intestinal Villus Microvasculature
Zhu, Yunwen (2026)
2026
Bachelor's Programme in Science and Engineering
Lääketieteen ja terveysteknologian tiedekunta - Faculty of Medicine and Health Technology
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2026-05-18
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202605135550
https://urn.fi/URN:NBN:fi:tuni-202605135550
Tiivistelmä
Micro-CT enables three-dimensional imaging of small-intestinal villus microvasculature, but vessel segmentations produced from such data often contain discontinuities. These breaks can limit downstream analyses such as skeletonization, centerline extraction, branch counting, and graph-based morphometry.
This thesis investigates an interpretable post-processing pipeline for reconnecting fragmented vessel segments in segmented 3D micro-CT data of small-intestinal villi. The method operates on an existing binary vessel labelmap. It combines skeleton-based endpoint detection, geometric candidate filtering, Frangi vesselness-based image support, and minimum-cost path reconstruction. Accepted paths are inserted back into the binary vessel mask to improve local connectivity.
The method was evaluated mainly by qualitative visual inspection, supported by pipeline-level indicators such as the number of endpoints, connected components, and accepted reconnections. Three parameter regimes were tested. After small-radius morphological closing, the initial skeleton contained 615 endpoints and 176 connected components. The standard setting accepted 160 reconnections and reduced the number of endpoints to 364 and connected components to 16. Local examples showed both successful reconnections and false lateral bridges.
The results suggest that vessel reconnection in 3D micro-CT data can be approached as a combined topological and image-guided optimization problem. However, missed gaps, false connections, and parameter sensitivity remain important limitations.
This thesis investigates an interpretable post-processing pipeline for reconnecting fragmented vessel segments in segmented 3D micro-CT data of small-intestinal villi. The method operates on an existing binary vessel labelmap. It combines skeleton-based endpoint detection, geometric candidate filtering, Frangi vesselness-based image support, and minimum-cost path reconstruction. Accepted paths are inserted back into the binary vessel mask to improve local connectivity.
The method was evaluated mainly by qualitative visual inspection, supported by pipeline-level indicators such as the number of endpoints, connected components, and accepted reconnections. Three parameter regimes were tested. After small-radius morphological closing, the initial skeleton contained 615 endpoints and 176 connected components. The standard setting accepted 160 reconnections and reduced the number of endpoints to 364 and connected components to 16. Local examples showed both successful reconnections and false lateral bridges.
The results suggest that vessel reconnection in 3D micro-CT data can be approached as a combined topological and image-guided optimization problem. However, missed gaps, false connections, and parameter sensitivity remain important limitations.
Kokoelmat
- Kandidaatintutkielmat [11997]