Varying pixel resolution significantly improves deep learning-based carotid plaque histology segmentation

Area of Science:

• Cardiovascular Pathology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Carotid plaques, composed of lipids, calcium, and debris, can rupture and cause strokes.
• Plaque composition and mechanical properties influence rupture risk.
• Automating histopathological analysis of carotid plaques is crucial for correlating histology with in vivo imaging and optimizing patient treatment.

Purpose of the Study:

• To develop and validate a deep learning model for automated segmentation of carotid plaque components.
• To investigate the impact of varied pixel resolution on the model's performance.
• To improve the accuracy of identifying lipid and calcified regions in carotid plaques.

Main Methods:

• Utilized Mask R-CNN deep learning model trained on 1944 regions of interest from 323 whole slide images.
• Varied pixel resolution of histology images ([Formula: see text] to [Formula: see text]) to provide contextual information.
• Compared performance against training with standard patches.

Main Results:

• Achieved a [Formula: see text] increase in pixel accuracy compared to training with patches.
• Obtained high F1 scores: [Formula: see text] for calcified regions, [Formula: see text] for lipid core, and [Formula: see text] for fibrous regions.
• Demonstrated qualitative ability to predict lumen; hemorrhage was excluded due to limited data.

Conclusions:

• Deep learning models trained on histology images with varied pixel resolution can accurately segment carotid plaque components.
• This approach enhances the potential for correlating histopathology with in vivo imaging for better stroke risk stratification.
• Further development could refine plaque analysis and inform clinical interventions.