A Semi-Automatic Magnetic Resonance Imaging Annotation Algorithm Based on Semi-Weakly Supervised Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate annotation of magnetic resonance imaging (MRI) images is crucial for deep learning-based segmentation.
• Current semi-automatic annotation methods struggle with insufficient segmentation labels, leading to poor pre-annotation performance.
• There is a need for efficient and effective semi-automatic annotation techniques to reduce the manual effort in MRI segmentation.

Purpose of the Study:

• To propose a novel semi-automatic MRI annotation algorithm utilizing semi-weakly supervised learning.
• To enhance pre-annotation performance in scenarios with limited segmentation labels.
• To improve the contribution of individual segmentation labels to the pre-annotation model's performance.

Main Methods:

• Developed a semi-weakly supervised learning segmentation algorithm leveraging sparse labels.
• Integrated semi-supervised and weakly supervised learning techniques.
• Implemented an iterative annotation strategy based on active learning to maximize label contribution.

Main Results:

• The proposed algorithm demonstrated equivalent pre-annotation performance compared to fully supervised methods, even with significantly fewer segmentation labels.
• Experimental results on public MRI datasets validated the algorithm's effectiveness.
• The approach successfully addressed the challenge of poor pre-annotation performance with insufficient labels.

Conclusions:

• The proposed semi-automatic MRI annotation algorithm based on semi-weakly supervised learning is effective in improving pre-annotation performance with limited data.
• The integration of semi-supervised, weakly supervised, and active learning strategies offers a robust solution for efficient MRI image annotation.
• This method significantly reduces the dependency on extensive labeled data for accurate MRI segmentation.