ResearchHealthcare

Transformer models for coronary artery wall segmentation in intravascular ultrasound

A review of transformer and attention-based deep learning architectures for outlining coronary artery walls in IVUS images, and where they outperform earlier networks.

The challenge

Intravascular ultrasound shows the inside of coronary arteries during angioplasty, but outlining the vessel wall and plaque by hand is slow and varies between readers. Accurate automatic segmentation is the foundation for measuring plaque burden.

Our approach

  1. Survey convolutional, attention-based and transformer architectures applied to IVUS segmentation.
  2. Compare their design choices, datasets and evaluation metrics.
  3. Identify open problems, including bias, generalisation and clinical validation.

The solution

  • Structured comparison of segmentation architectures
  • Recommendations for clinical validation

Results

  • Published in Diagnostics (MDPI), 2025, volume 15, article 848 — open access.
  • Guides the choice of architectures for BigHealth’s cardiovascular imaging work.

Publication

Kumari, Katiyar, Bhagawati, Maindarkar et al. (2025). Diagnostics, 15(7), 848. View source

Have a similar challenge?

Tell us about your data and your goal. We reply within one working day.

Talk to us