Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Accelerated Intravascular Ultrasound Imaging using Deep Reinforcement Learning

Tristan S. W. Stevens, Nishith Chennakeshava, Frederik J. de Bruijn, Martin Pekař, Ruud J. G. van Sloun

TL;DR

The paper addresses accelerating intravascular ultrasound (IVUS) imaging under a tight information bottleneck from a limited catheter-wide bandwidth. It introduces AiVUS, a deep reinforcement learning framework that learns a per-frame adaptive acquisition policy using actor-critic methods and Gumbel top-$K$ sampling to select $K$ measurements from $N$ possible transducer element pairs in one frame. By modeling the process as a POMDP and conditioning the current frame's sampling on prior reconstructions, AiVUS achieves higher image quality under data-rate constraints, demonstrated across simulated wire targets, wire phantoms, and in-vivo porcine data, with improvements measured in MSE, MAE, PSNR, and SSIM. The results indicate a practical route to faster IVUS with preserved diagnostic detail, and the framework can be extended to control other ultrasound transmit settings and modalities.

Abstract

Intravascular ultrasound (IVUS) offers a unique perspective in the treatment of vascular diseases by creating a sequence of ultrasound-slices acquired from within the vessel. However, unlike conventional hand-held ultrasound, the thin catheter only provides room for a small number of physical channels for signal transfer from a transducer-array at the tip. For continued improvement of image quality and frame rate, we present the use of deep reinforcement learning to deal with the current physical information bottleneck. Valuable inspiration has come from the field of magnetic resonance imaging (MRI), where learned acquisition schemes have brought significant acceleration in image acquisition at competing image quality. To efficiently accelerate IVUS imaging, we propose a framework that utilizes deep reinforcement learning for an optimal adaptive acquisition policy on a per-frame basis enabled by actor-critic methods and Gumbel top-$K$ sampling.

Accelerated Intravascular Ultrasound Imaging using Deep Reinforcement Learning

TL;DR

The paper addresses accelerating intravascular ultrasound (IVUS) imaging under a tight information bottleneck from a limited catheter-wide bandwidth. It introduces AiVUS, a deep reinforcement learning framework that learns a per-frame adaptive acquisition policy using actor-critic methods and Gumbel top- sampling to select measurements from possible transducer element pairs in one frame. By modeling the process as a POMDP and conditioning the current frame's sampling on prior reconstructions, AiVUS achieves higher image quality under data-rate constraints, demonstrated across simulated wire targets, wire phantoms, and in-vivo porcine data, with improvements measured in MSE, MAE, PSNR, and SSIM. The results indicate a practical route to faster IVUS with preserved diagnostic detail, and the framework can be extended to control other ultrasound transmit settings and modalities.

Abstract

Intravascular ultrasound (IVUS) offers a unique perspective in the treatment of vascular diseases by creating a sequence of ultrasound-slices acquired from within the vessel. However, unlike conventional hand-held ultrasound, the thin catheter only provides room for a small number of physical channels for signal transfer from a transducer-array at the tip. For continued improvement of image quality and frame rate, we present the use of deep reinforcement learning to deal with the current physical information bottleneck. Valuable inspiration has come from the field of magnetic resonance imaging (MRI), where learned acquisition schemes have brought significant acceleration in image acquisition at competing image quality. To efficiently accelerate IVUS imaging, we propose a framework that utilizes deep reinforcement learning for an optimal adaptive acquisition policy on a per-frame basis enabled by actor-critic methods and Gumbel top- sampling.
Paper Structure (11 sections, 8 equations, 3 figures, 1 table)

This paper contains 11 sections, 8 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. IVUS Signal Acquisition
  4. Subsampling
  5. Reinforcement Learning for Adaptive Subsampling
  6. Experiments
  7. Simulated Wire Target
  8. Wire Phantom
  9. In-vivo data
  10. Results
  11. Conclusion

Figures (3)

  • Figure 1: Block diagram of the perception-action-learning loop of the RL agent in an IVUS imaging setting.
  • Figure 2: Six successive wire phantom frames constructed using AiVUS. The agent's action is displayed in the top row, where the circle represents the elements in the transducer array. Actions are displayed such that black represents 1 receiving element and white $A$ receiving elements, for each transmitting element $\in \left\{1, \ldots, E\right\}$.
  • Figure 3: SSIM performance on the in-vivo test data for both learned (AiVUS) and random sampling strategies. Average and standard deviations are computed over five seeds.