Automatic microchannel detection using deep learning in intravascular optical coherence tomography images.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Microchannels in atherosclerotic plaques are key indicators of vulnerability, rupture, and hemorrhage, driving disease progression.
• Accurate detection of microchannels is crucial for assessing plaque stability and guiding treatment strategies.

Purpose of the Study:

• To develop and validate a fully-automated deep learning method for detecting microchannels in intravascular optical coherence tomography (IVOCT) images.
• To assess the performance and clinical potential of the automated microchannel detection system.

Main Methods:

• A deep learning approach utilizing the DeepLab-v3 plus model with an Xception backbone was employed for microchannel candidate identification.
• Pre-processing steps included guidewire removal, lumen segmentation, pixel-shifting, and noise filtering to enhance raw IVOCT images.
• A convolutional neural network (CNN) classifier was used to differentiate true microchannels from false positives, improving overall accuracy.

Main Results:

• The automated method achieved high segmentation performance with a Dice coefficient of 0.811, 92.4% sensitivity, and 99.9% specificity.
• Pre-processing and data augmentation significantly improved segmentation results, while the CNN classifier effectively reduced false positives.
• The system demonstrated excellent performance, missing only 3% of microchannel-containing frames and exhibiting no false positives.

Conclusions:

• The developed automated deep learning method offers a highly objective, repeatable, and comprehensive tool for evaluating vulnerable plaques.
• This technique shows significant promise for both research applications and clinical integration in cardiovascular diagnostics and treatment monitoring.