Detection of Brain Mid-Sagittal Plane Based on Progressive Semi-Supervised Pixel Classification Algorithm.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computer Vision

Background:

• Automatic mid-sagittal plane (MSP) detection is crucial for brain imaging analysis, including symmetry analysis and morphometry.
• Current fully supervised learning (SL) methods are limited by the cost and availability of annotated data, hindering model generalization.

Purpose of the Study:

• To develop an improved MSP detection framework that enhances generalization across various brain image datasets.
• To overcome the limitations of existing methods by leveraging unlabeled data and incorporating robustness features.

Main Methods:

• Proposed a Progressive Semi-supervised Learning (PSSL) method utilizing morphological characteristics of the MSP for continuous learning from unlabeled data.
• Integrated a 3D neighborhood-based correction mechanism to enhance the model's fault tolerance.

Main Results:

• The PSSL framework demonstrated significant performance improvements using large amounts of unlabeled data.
• The integrated correction mechanism provided robustness against potential errors in detection.
• Extensive validation on five datasets with millions of brain sections confirmed superior performance over state-of-the-art methods.

Conclusions:

• The proposed framework effectively improves the generalization of MSP detection models.
• This approach offers a more robust and accurate solution for midline detection in brain imaging, applicable to large-scale datasets.