Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fitting tree model with CNN and geodesics to track vesselsand application to Ultrasound Localization Microscopy data

Théo Bertrand, Laurent D. Cohen

TL;DR

This work focuses on detecting the important landmarks in the vascular network and representing vessels as the edges in some minimal distance tree graph, using geodesic methods relevant to the detection of vessels and their geometry to make use of the space of positions and orientations.

Abstract

Segmentation of tubular structures in vascular imaging is a well studied task, although it is rare that we try to infuse knowledge of the tree-like structure of the regions to be detected. Our work focuses on detecting the important landmarks in the vascular network (via CNN performing both localization and classification of the points of interest) and representing vessels as the edges in some minimal distance tree graph. We leverage geodesic methods relevant to the detection of vessels and their geometry, making use of the space of positions and orientations so that 2D vessels can be accurately represented as trees. We build our model to carry tracking on Ultrasound Localization Microscopy (ULM) data, proposing to build a good cost function for tracking on this type of data. We also test our framework on synthetic and eye fundus data. Results show that scarcity of well annotated ULM data is an obstacle to localization of vascular landmarks but the Orientation Score built from ULM data yields good geodesics for tracking blood vessels.

Fitting tree model with CNN and geodesics to track vesselsand application to Ultrasound Localization Microscopy data

TL;DR

This work focuses on detecting the important landmarks in the vascular network and representing vessels as the edges in some minimal distance tree graph, using geodesic methods relevant to the detection of vessels and their geometry to make use of the space of positions and orientations.

Abstract

Segmentation of tubular structures in vascular imaging is a well studied task, although it is rare that we try to infuse knowledge of the tree-like structure of the regions to be detected. Our work focuses on detecting the important landmarks in the vascular network (via CNN performing both localization and classification of the points of interest) and representing vessels as the edges in some minimal distance tree graph. We leverage geodesic methods relevant to the detection of vessels and their geometry, making use of the space of positions and orientations so that 2D vessels can be accurately represented as trees. We build our model to carry tracking on Ultrasound Localization Microscopy (ULM) data, proposing to build a good cost function for tracking on this type of data. We also test our framework on synthetic and eye fundus data. Results show that scarcity of well annotated ULM data is an obstacle to localization of vascular landmarks but the Orientation Score built from ULM data yields good geodesics for tracking blood vessels.
Paper Structure (14 sections, 2 equations, 4 figures)

This paper contains 14 sections, 2 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Detecting vascular landmarks
  3. ULM Data
  4. Training
  5. Finding appropriate geodesics
  6. Geodesics for vessel tracking
  7. Clustering landmarks using the geodesic graph
  8. Linking landmarks through the geodesic graph
  9. Results and Discussion
  10. Synthetic data
  11. Eye fundus image
  12. Rat Brain ULM data
  13. Discussion
  14. Conclusion and further works

Figures (4)

  • Figure 1: Top : Patches are made from a high resolution image by cropping patches taken uniformly from each brain half. Left : eye fundus image overlayed with heatmap of vascular landmarks. Right : ULM image overlayed with heatmap of vascular landmarks.
  • Figure 2: Geodesic graph on half of the synthetic validation image, with $N_{\theta} = 128$. Left image shows detected landmarks points and the geodesics linking them, right image shows the input image.
  • Figure 3: Geodesic tracking performed on two validation images from the DRIVE and IOSTAR dataset, with $N_{\theta} = 64.$ Big red points are the detected landmarks and curves are the selected tree structures.
  • Figure 4: Geodesic graph on patches cropped from the ULM validation dataset (taken respectively from left and right parts of rat brain), with $N_{\theta}=64$. Big red points are the detected landmarks and curves are the selected tree structures.