FDCT: Fusion-Guided Dual-View Consistency Training for semi-supervised tissue segmentation on MRI

Area of Science:

• Medical Imaging Analysis
• Machine Learning for Healthcare
• Computational Anatomy

Background:

• Accurate Magnetic Resonance Imaging (MRI) tissue segmentation is crucial for clinical diagnosis and treatment planning.
• Existing segmentation models often lack generality, performing poorly on diverse MRI tasks.
• The scarcity of labeled data presents a significant hurdle in developing robust segmentation models.

Purpose of the Study:

• To develop a universal, semi-supervised method for multi-task MRI tissue segmentation.
• To enhance the generality and robustness of segmentation models.
• To mitigate the challenges associated with acquiring large labeled datasets.

Main Methods:

• Proposed Fusion-Guided Dual-View Consistency Training (FDCT) for semi-supervised MRI tissue segmentation.
• Employed a single-encoder dual-decoder architecture with dual-view image input for view-level predictions.
• Introduced a fusion module to generate image-level pseudo-labels and a Soft-label Boundary Optimization Module (SBOM) for improved boundary accuracy.

Main Results:

• FDCT demonstrated accurate and robust tissue segmentation capabilities across multiple MRI tasks.
• The method effectively alleviated the problem of insufficient labeled data.
• Extensive experiments on three MRI datasets confirmed the superiority of FDCT over state-of-the-art semi-supervised methods.

Conclusions:

• FDCT offers a universal and effective solution for semi-supervised multi-task MRI tissue segmentation.
• The proposed approach significantly improves segmentation accuracy and data efficiency.
• FDCT represents a promising advancement in medical image analysis, aiding clinical decision-making.