Carotid Plaque Segmentation in Ultrasound Images Using a Mask R-CNN

TL;DR

This study refines a Mask R-CNN model to automatically segment carotid atherosclerotic plaques in ultrasound images, aiding in diagnosis and treatment planning.

Contribution

The paper adapts and evaluates a Mask R-CNN model for carotid plaque segmentation using ultrasound images, demonstrating improved accuracy with model ensemble techniques.

Findings

Highest Dice score of 0.74 with 2 prediction regions

Ensemble analysis improved scores to 0.76

Model performance varies with plaque presentation complexity

Abstract

Background: Ultrasound imaging plays a pivotal role in diagnosing carotid atherosclerosis, a significant precursor to cardiovascular and cerebrovascular diseases and events. This non-invasive modality provides real-time, high-resolution images, allowing clinicians to assess atherosclerotic plaques in the carotid arteries without invasive procedures. Purpose: In this study, we present the refinement of a Mask R-CNN model initially designed for carotid lumen detection to automatically generate bounding boxes enclosing atherosclerotic plaque for segmentation to assist in our ultrasound elastography workflow. Methods: We utilize a PyTorch torchvision implementation of the Mask R-CNN for carotid plaque detection. Our dataset consists of 118 severe stenotic carotid plaques from presenting patients, clinically indicated for a carotid endarterectomy. Due to the variability of plaque presentation in the dataset, a multitude of different R-CNN models were observed to have varying results based on the allowed number of prediction regions. An overview analysis looking at shared predictions from these models showed a slight improvement compared to the individual model results Results: Evaluation metrics such as Dice similarity coefficient and intersection over Union are employed. The model trained with 5 maximum prediction regions and tested with 2 maximum prediction regions produced the highest individual accuracy with a Dice score of 0.74 and intersection over union of 0.61. A filtered combined analysis of all the models demonstrated a slight increase in performance with scores of 0.76 and 0.61 respectively. Conclusion: Due to the significant variation in plaque presentation and types amongst presenting patients, the accuracy of the Plaque Mask R-CNN network would benefit from the incorporation of additional patient datasets to incorporate increased variation into the training dataset.