WAL-Net: Weakly supervised auxiliary task learning network for carotid plaques classification

TL;DR

This work tackles carotid plaque ultrasound classification by exploiting the correlation between segmentation and classification without requiring segmentation labels. It introduces WAL-Net, an end-to-end multitask framework with a shared encoder, a weakly supervised segmentation head guided by a Pseudo Mask Generation Module and a Region of Interest Cropping Module to reinforce the classification task. On a dataset of 1,270 images, WAL-Net achieves a ~1.3% absolute accuracy gain over a robust baseline, with a notable 3.3% improvement for mixed-echoic plaques, demonstrating that weakly supervised auxiliary segmentation can enhance primary classification performance while reducing annotation burden. The approach holds potential for broader application in medical imaging where auxiliary tasks can be leveraged without extensive labeling.

Abstract

The classification of carotid artery ultrasound images is a crucial means for diagnosing carotid plaques, holding significant clinical relevance for predicting the risk of stroke. Recent research suggests that utilizing plaque segmentation as an auxiliary task for classification can enhance performance by leveraging the correlation between segmentation and classification tasks. However, this approach relies on obtaining a substantial amount of challenging-to-acquire segmentation annotations. This paper proposes a novel weakly supervised auxiliary task learning network model (WAL-Net) to explore the interdependence between carotid plaque classification and segmentation tasks. The plaque classification task is primary task, while the plaque segmentation task serves as an auxiliary task, providing valuable information to enhance the performance of the primary task. Weakly supervised learning is adopted in the auxiliary task to completely break away from the dependence on segmentation annotations. Experiments and evaluations are conducted on a dataset comprising 1270 carotid plaque ultrasound images from Wuhan University Zhongnan Hospital. Results indicate that the proposed method achieved an approximately 1.3% improvement in carotid plaque classification accuracy compared to the baseline network. Specifically, the accuracy of mixed-echoic plaques classification increased by approximately 3.3%, demonstrating the effectiveness of our approach.

Figures (6)

• Figure 1: Overview of the proposed WAL-Net. Bottom right: Structure of the ROI Cropping Module.
• Figure 2: Structure of the Pseudo mask Generation Module.
• Figure 3: The preprocessing pipeline for carotid artery plaque ultrasound images. (a): The original ultrasound images of the three types of plaques: hyperechoic plaque, hypoechoic plaque, and mixed-echoic plaque. (b): The regions of interest in the ultrasound images of the three types of plaques. (c): The ultrasound images of the three types of plaques after being uniformly resized.
• Figure 4: ROC curves for WAL-Net and baseline.
• Figure 5: Confusion matrix for WAL-Net and baseline.