AMLP: Adjustable Masking Lesion Patches for Self-Supervised Medical Image Segmentation.

Area of Science:

• Medical image analysis
• Computer vision
• Machine learning

Background:

• Self-supervised masked image modeling (MIM) shows promise for natural images but struggles with complex medical images.
• High, fixed masking ratios in MIM can obscure critical background information in medical scans.
• Distinct contour features in medical images present unique challenges for standard MIM approaches.

Purpose of the Study:

• To develop a novel self-supervised framework, Adjustable Masking Lesion Patches (AMLP), for improved medical image segmentation.
• To address the limitations of applying MIM to complex medical images, particularly in lesion detection.
• To enhance the accuracy of lesion segmentation by focusing on relevant image patches and reconstruction difficulty.

Main Methods:

• Proposed Adjustable Masking Lesion Patches (AMLP) framework utilizing Masked Patch Selection (MPS) for lesion-rich patch identification.
• Introduced Relative Reconstruction Loss (RRL) to improve the learning of difficult-to-reconstruct lesion patches.
• Implemented Category Consistency Loss (CCL) for refined patch categorization and Category Consistency Loss (CCL) for enhanced lesion-background differentiation.
• Developed an Adjustable Masking Ratio (AMR) strategy to progressively increase masking during training, expanding learnable information.

Main Results:

• AMLP demonstrated superior performance compared to state-of-the-art self-supervised methods on two medical segmentation datasets.
• The proposed MPS, RRL, CCL, and AMR strategies effectively addressed challenges in applying masked modeling to medical images.
• AMLP successfully captured intricate lesion details crucial for accurate medical image segmentation.

Conclusions:

• The AMLP framework significantly advances self-supervised learning for medical image segmentation.
• AMLP offers a robust solution for analyzing complex medical images and accurately segmenting lesions.
• This approach holds potential for improving diagnostic accuracy and clinical decision-making in medical imaging.