The paper introduces DCL-Net, a novel approach for semi-supervised multi-organ segmentation using dual contrastive learning. It combines global and local contrastive learning to improve feature representations and achieve better segmentation results. Experimental results on ACDC and RC-OARs datasets show the effectiveness of the proposed method compared to state-of-the-art techniques.
The content discusses the challenges of multi-organ segmentation in medical imaging due to limited annotated data availability. It presents a two-stage approach involving global and local contrastive learning to enhance feature extraction and improve segmentation accuracy. The methodology includes innovative strategies like mask center computation and memory bank maintenance for efficient representation learning.
Key components of the DCL-Net model are explained, including similarity-guided global contrastive learning in Stage I and organ-aware local contrastive learning in Stage II. The paper details objective functions, training procedures, dataset descriptions, evaluation metrics, and comparisons with other state-of-the-art methods.
Experimental results demonstrate the superior performance of DCL-Net over existing methods in terms of Dice coefficient and Jaccard Index on ACDC and RC-OARs datasets. Visualization comparisons highlight the accuracy and effectiveness of the proposed approach in multi-organ segmentation tasks.
toiselle kielelle
lähdeaineistosta
arxiv.org
Tärkeimmät oivallukset
by Lu Wen,Zheng... klo arxiv.org 03-07-2024
https://arxiv.org/pdf/2403.03512.pdfSyvällisempiä Kysymyksiä