Carotid Lumen Diameter and Intima-Media Thickness Measurement Via Boundary-Guided Pseudo-Labeling.

Shimeng Yang,Teng Li,Yinping Lv,Yi Xia,Shuo Li

DOI: https://doi.org/10.1109/lsp.2023.3292028

2023-01-01

IEEE Signal Processing Letters

Abstract:The carotid lumen diameter (CALD) and intima-media thickness (CIMT) are essential indicators for diagnosing and treating cardiovascular disease. Existing supervised methods rely on extensive labeled data, which is time-consuming and labor-intensive. This letter presents a boundary-guided pseudo-labeling (BGPL) method, which adopts prior anatomical knowledge to generate and select reliable pseudo-labels for unlabeled data to boost measurement performance. To improve the quality of pseudo-labels, we propose a fine self-attention (FSA) module and a boundary attention module (BAM) to force the feature extractor to highlight boundary information. The FSA maintains internal resolution and enhances non-linearity. The BAM injects boundary heatmaps from the pre-trained boundary regressor into the feature extractor. We specifically provide an alternate training strategy to increase the quality of pseudo-labels further. We evaluate the proposed method on challenging carotid ultrasound datasets. Experiments show that the proposed method outperforms the existing state-of-the-art algorithm.