Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TractOracle: towards an anatomically-informed reward function for RL-based tractography

Antoine Théberge, Maxime Descoteaux, Pierre-Marc Jodoin

TL;DR

Tractography remains ill-posed due to reliance on local diffusion cues, which can yield spurious pathways. The authors introduce TractOracle, an RL-based framework that couples streamline tracking with an anatomically informed reward derived from a transformer-based TractOracle-Net that scores streamlines for plausibility. The approach yields improved true positive rates and reduced false positives across synthetic and in-vivo data, achieving state-of-the-art tractogram quality in both in-silico and in-vivo evaluations. By enabling simultaneous tracking and filtering with an anatomical prior, this method has the potential to enhance clinical applicability of tractography by reducing spurious connections while preserving genuine white-matter pathways.

Abstract

Reinforcement learning (RL)-based tractography is a competitive alternative to machine learning and classical tractography algorithms due to its high anatomical accuracy obtained without the need for any annotated data. However, the reward functions so far used to train RL agents do not encapsulate anatomical knowledge which causes agents to generate spurious false positives tracts. In this paper, we propose a new RL tractography system, TractOracle, which relies on a reward network trained for streamline classification. This network is used both as a reward function during training as well as a mean for stopping the tracking process early and thus reduce the number of false positive streamlines. This makes our system a unique method that evaluates and reconstructs WM streamlines at the same time. We report an improvement of true positive ratios by almost 20\% and a reduction of 3x of false positive ratios on one dataset and an increase between 2x and 7x in the number true positive streamlines on another dataset.

TractOracle: towards an anatomically-informed reward function for RL-based tractography

TL;DR

Tractography remains ill-posed due to reliance on local diffusion cues, which can yield spurious pathways. The authors introduce TractOracle, an RL-based framework that couples streamline tracking with an anatomically informed reward derived from a transformer-based TractOracle-Net that scores streamlines for plausibility. The approach yields improved true positive rates and reduced false positives across synthetic and in-vivo data, achieving state-of-the-art tractogram quality in both in-silico and in-vivo evaluations. By enabling simultaneous tracking and filtering with an anatomical prior, this method has the potential to enhance clinical applicability of tractography by reducing spurious connections while preserving genuine white-matter pathways.

Abstract

Reinforcement learning (RL)-based tractography is a competitive alternative to machine learning and classical tractography algorithms due to its high anatomical accuracy obtained without the need for any annotated data. However, the reward functions so far used to train RL agents do not encapsulate anatomical knowledge which causes agents to generate spurious false positives tracts. In this paper, we propose a new RL tractography system, TractOracle, which relies on a reward network trained for streamline classification. This network is used both as a reward function during training as well as a mean for stopping the tracking process early and thus reduce the number of false positive streamlines. This makes our system a unique method that evaluates and reconstructs WM streamlines at the same time. We report an improvement of true positive ratios by almost 20\% and a reduction of 3x of false positive ratios on one dataset and an increase between 2x and 7x in the number true positive streamlines on another dataset.
Paper Structure (9 sections, 7 equations, 2 figures, 3 tables)

This paper contains 9 sections, 7 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Preliminaries and proposed method
  3. Implementation
  4. Experiments and results
  5. Datasets and tools
  6. TractOracle-Net performance
  7. TractOracle in-silico performance
  8. TractOracle in-vivo performance
  9. Discussion and conclusion

Figures (2)

  • Figure 1: (a) TractOracle: an RL system in which the environment sends streamlines to TractOracle-Net as they are being tracked to score their anatomical plausibility. The scores are used to reward the agent and stop the tracking process when streamlines diverge into an implausible shape. TractOracle-RL then uses the reward function to predict anatomically-informed tractograms. (b) TractOracle-Net scores along valid and invalid cortico-spinal tracts. Streamlines correctly terminating in the motor cortex get a high plausibility score ( red); implausible streamlines diverging towards the corpus callosum get a low ( blue) score.
  • Figure 2: Visualization of the occipital part of the corpus callosum (1st row), the left cingulum (2nd row), the left-right uncinate fasciculus (3rd row) and the left-right parieto-occipito pontine tracts (4th row) from subject 1006 of the Tractoinferno dataset (reference) and as reconstructed by all methods considered in this work.