Domain Adapted Model for In Vivo Intravascular Ultrasound Tissue Characterization

S. Conjeti, A. G. Roy, D. Sheet, S. Carlier, T. Syeda-Mahmood, N. Navab, A. Katouzian

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

Abstract

Intravascular ultrasound (IVUS) is a real-time cross-sectional imaging modality deployed in interventional cardiology for assessment of the extent of atherosclerosis. Visual reading of IVUS pull-backs is subject to inter- and intra-observer variability in reporting of vulnerable plaques causing myocardial infraction. In vivo IVUS tissue characterization (TC) aims at augmenting information about the constituent tissues beyond features visible in the log-compressed B-Mode scan by effectively leveraging characteristic ultrasonic backscattered signals acquired during live intervention. The co-located heterogeneity of biological tissues constituting the plaque, the presence of flowing blood and vessel dynamics openly challenge in vivo TC. As a solution, we introduce a framework that first uses a decision forest based classifier that learns to perform TC using tissue specific ultrasonic statistical physics and signal confidence features, from labeled data acquired under controlled in vitro conditions. Next, we adapt this in vitro trained classifier to work under in vivo settings through a novel error-correcting hierarchical transfer relaxation scheme for domain adaptation with few labeled samples. This effectively compensates for the shift in statistical features between in vitro and in vivo settings owed to the presence of flowing blood and vessel dynamic movements. Experiments reveal the ability of the framework to estimate constituents of the plaque reliably under both in vitro and in vivo settings. This framework can be leveraged for promising clinically applications requiring TC and to perform domain adaptation in the presence of few labeled samples.

Original languageEnglish
Title of host publicationComputing and Visualization for Intravascular Imaging and Computer-Assisted Stenting
PublisherElsevier Inc.
Pages157-181
Number of pages25
ISBN (Electronic)9780128110195
ISBN (Print)9780128110188
DOIs
StatePublished - 2017

Keywords

  • Domain adaptation
  • Intravascular ultrasound
  • Random forests
  • Tissue characterization

Fingerprint

Dive into the research topics of 'Domain Adapted Model for In Vivo Intravascular Ultrasound Tissue Characterization'. Together they form a unique fingerprint.

Cite this