Accelerated Intravascular Ultrasound Imaging using Deep Reinforcement Learning

  • 2022-01-24 08:33:21
  • Tristan S. W. Stevens, Nishith Chennakeshava, Frederik J. de Bruijn, Martin Pekaƙ, Ruud J. G. van Sloun
  • 3

Abstract

Intravascular ultrasound (IVUS) offers a unique perspective in the treatmentof vascular diseases by creating a sequence of ultrasound-slices acquired fromwithin the vessel. However, unlike conventional hand-held ultrasound, the thincatheter only provides room for a small number of physical channels for signaltransfer from a transducer-array at the tip. For continued improvement of imagequality and frame rate, we present the use of deep reinforcement learning todeal with the current physical information bottleneck. Valuable inspiration hascome from the field of magnetic resonance imaging (MRI), where learnedacquisition schemes have brought significant acceleration in image acquisitionat competing image quality. To efficiently accelerate IVUS imaging, we proposea framework that utilizes deep reinforcement learning for an optimal adaptiveacquisition policy on a per-frame basis enabled by actor-critic methods andGumbel top-$K$ sampling.

 

Quick Read (beta)

loading the full paper ...