A contrastive consistency semi-supervised left atrium segmentation model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Accurate left atrium (LA) segmentation is crucial for diagnosing and treating atrial fibrillation.
• Current supervised deep learning methods require extensive labeled data, which is difficult to obtain.
• Automated segmentation methods are needed to reduce clinical workload and improve efficiency.

Purpose of the Study:

• To develop a semi-supervised deep learning framework for efficient and accurate LA segmentation.
• To leverage unlabeled data to overcome the limitations of data scarcity in supervised learning.
• To improve the clinical diagnosis and therapy of atrial fibrillation through automated LA segmentation.

Main Methods:

• A semi-supervised framework combining a segmentation model and a classification model was proposed.
• A contrastive consistency loss function was developed to enhance feature representation using class-vectors.
• Class-vectors from labeled data were used as references for unlabeled data to mitigate prediction unreliability.

Main Results:

• The proposed framework achieved state-of-the-art performance on a public LA dataset.
• Achieved a Dice Score of 89.81%, Jaccard of 81.64%, 95% Hausdorff distance of 7.15 mm, and Average Surface Distance of 1.82 mm.
• Demonstrated superior performance compared to existing state-of-the-art models across all evaluated metrics.

Conclusions:

• The developed semi-supervised framework offers a highly effective solution for automated LA segmentation.
• This approach significantly reduces the reliance on large labeled datasets, making it more practical for clinical use.
• The framework shows potential for substantial contributions to assisting atrial fibrillation therapy and improving patient outcomes.