Automatic lumen and anatomical layers segmentation in IVOCT images using meta learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Research

Background:

• Automated analysis of intravascular optical coherence tomography (IVOCT) images is crucial for assessing vessel health and monitoring coronary artery disease.
• Deep learning models typically require extensive annotated datasets, which are challenging to acquire in medical imaging.

Purpose of the Study:

• To develop an automatic method for segmenting vessel layers (lumen, intima, media, adventitia) in IVOCT images using meta-learning.
• To address the challenge of limited annotated data in medical image analysis for cardiovascular applications.

Main Methods:

• A meta-learning framework utilizing a bi-level gradient strategy to train a meta-learner.
• Development of a Claw-type network and a contrast consistency loss for enhanced meta-knowledge acquisition.
• Application of the method to cardiovascular IVOCT datasets.

Main Results:

• The proposed meta-learning method achieved state-of-the-art performance in segmenting anatomical layers.
• Demonstrated effectiveness in extracting lumen, intima, media, and adventitia surfaces with few annotated samples.
• Validated on two cardiovascular IVOCT datasets.

Conclusions:

• Meta-learning offers a viable solution for automated vessel layer segmentation in IVOCT images, overcoming data limitations.
• The developed method shows significant potential for improving the assessment and monitoring of coronary artery disease.
• This approach facilitates more efficient and accurate analysis of cardiovascular structures from IVOCT data.