Coronary artery calcification segmentation with sparse annotations in intravascular OCT: Leveraging self-supervised...

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Coronary artery calcification (CAC) assessment is vital for atherosclerosis progression and percutaneous coronary intervention (PCI) planning.
• Intravascular Optical Coherence Tomography (OCT) provides detailed 3D evaluation of CAC for PCI optimization.
• Current deep learning models for OCT analysis require extensive high-quality labels, hindering practical application.

Purpose of the Study:

• To develop an annotation-efficient deep learning approach for segmenting CAC in intravascular OCT images.
• To leverage self-supervised learning and consistency regularization to overcome data labeling limitations.
• To improve the performance and efficiency of AI models in analyzing cardiovascular OCT data.

Main Methods:

• Utilized a transformer encoder with a linear projection layer for self-supervised pre-training on unlabeled OCT data.
• Fine-tuned a transformer-based segmentation model using sparsely annotated OCT data and a contrast loss function.
• Employed a large dataset: 2,549,073 unlabeled OCT images for pre-training and 1,106,347 sparsely annotated images for fine-tuning and testing.

Main Results:

• The proposed method demonstrated superior performance compared to existing sparsely supervised techniques on both internal and external datasets.
• Extensive evaluations confirmed the approach's high annotation efficiency across full, partial, and sparse annotation scenarios.
• Achieved an 80% reduction in image labeling efforts while maintaining high performance.

Conclusions:

• The developed annotation-efficient method effectively segments CAC in intravascular OCT images.
• This approach significantly reduces the need for extensive manual labeling, making deep learning more practical for large-scale medical image analysis.
• The findings pave the way for faster development and deployment of AI tools in cardiovascular imaging.