BMAnet: Boundary Mining With Adversarial Learning for Semi-Supervised 2D Myocardial Infarction Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Accurate segmentation of myocardial infarction (MI) regions in cardiac MRI is crucial for diagnosis.
• Current fully supervised deep learning methods require extensive, time-consuming, and expensive cardiologist annotations.
• Limitations exist in existing segmentation techniques regarding boundary ambiguity and robustness.

Purpose of the Study:

• To develop an efficient semi-supervised deep learning model for myocardial infarction segmentation.
• To address the challenges of boundary ambiguity and limited labeled data in cardiac MRI segmentation.
• To improve the accuracy and robustness of automated MI region identification.

Main Methods:

• Proposed a novel semi-supervised learning framework comprising two models: a boundary mining model and an adversarial learning model.
• The boundary mining model enhances feature separation to resolve segmentation ambiguity.
• The adversarial learning model leverages unlabeled data through pseudo-supervision to boost model robustness.

Main Results:

• The proposed semi-supervised method achieved excellent results in myocardial infarction segmentation across six evaluation metrics.
• Experimental results demonstrated superior performance compared to existing state-of-the-art semi-supervised segmentation techniques.
• The integrated approach effectively improved segmentation accuracy and model robustness on an in-house dataset.

Conclusions:

• The developed semi-supervised approach offers a promising solution for automated myocardial infarction segmentation in cardiac MRI.
• This method reduces reliance on manual annotations, making automated diagnosis more efficient and cost-effective.
• The combination of boundary mining and adversarial learning enhances segmentation performance and generalizability.