Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep vessel segmentation based on a new combination of vesselness filters

Guillaume Garret, Antoine Vacavant, Carole Frindel

TL;DR

An innovative filter fusion method crafted to amplify the effectiveness of vessel segmentation models when the model’s learning is exposed to vessel-enhanced inputs is introduced.

Abstract

Vascular segmentation represents a crucial clinical task, yet its automation remains challenging. Because of the recent strides in deep learning, vesselness filters, which can significantly aid the learning process, have been overlooked. This study introduces an innovative filter fusion method crafted to amplify the effectiveness of vessel segmentation models. Our investigation seeks to establish the merits of a filter-based learning approach through a comparative analysis. Specifically, we contrast the performance of a U-Net model trained on CT images with an identical U-Net configuration trained on vesselness hyper-volumes using matching parameters. Our findings, based on two vascular datasets, highlight improved segmentations, especially for small vessels, when the model's learning is exposed to vessel-enhanced inputs.

Deep vessel segmentation based on a new combination of vesselness filters

TL;DR

An innovative filter fusion method crafted to amplify the effectiveness of vessel segmentation models when the model’s learning is exposed to vessel-enhanced inputs is introduced.

Abstract

Vascular segmentation represents a crucial clinical task, yet its automation remains challenging. Because of the recent strides in deep learning, vesselness filters, which can significantly aid the learning process, have been overlooked. This study introduces an innovative filter fusion method crafted to amplify the effectiveness of vessel segmentation models. Our investigation seeks to establish the merits of a filter-based learning approach through a comparative analysis. Specifically, we contrast the performance of a U-Net model trained on CT images with an identical U-Net configuration trained on vesselness hyper-volumes using matching parameters. Our findings, based on two vascular datasets, highlight improved segmentations, especially for small vessels, when the model's learning is exposed to vessel-enhanced inputs.
Paper Structure (17 sections, 3 figures, 2 tables)

This paper contains 17 sections, 3 figures, 2 tables.

Table of Contents

  1. Abstract
  2. Introduction
  3. Materials
  4. Data
  5. IRCAD
  6. Bullitt
  7. Vesselness filters
  8. Experimental setup
  9. Experiments
  10. Training
  11. Evaluation
  12. Results
  13. IRCAD
  14. Bullitt
  15. Conclusion
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Hyper-volume where filters are concatenated to the original scan, with each $Z_{i}$ depth corresponding to a different filter (Original, Frangi, Jerman, Sato, Zhang, Meijering, RORPO, respectively).
  • Figure 2: Signal loss for a large vessel. White: enhanced vessels; purple:ground truth; green:mask $M_{large}$.
  • Figure 3: Example of vessel segmentation for IRCAD (left) and Bullitt (right) datasets. Top to bottom: Ground-truth, U-Net's output using original images, and U-Net's output using our proposed hyper-volumes.