Diameter-based pseudo labeling for pathological image segmentation.

Area of Science:

• Medical image analysis
• Computational pathology
• Artificial intelligence in medicine

Background:

• Accurate pathological image segmentation is crucial for diagnosis and treatment planning.
• Weakly supervised methods offer a promising alternative to fully supervised approaches, reducing annotation burden.
• Leveraging clinical data like tumor diameter can provide valuable weak supervision signals.

Purpose of the Study:

• To propose a novel semi-and-weakly supervised method for pathological image segmentation.
• To utilize pre-recorded tumor long-diameter information as a form of weak supervision.
• To improve the accuracy of pathological image segmentation by generating high-quality pseudo-labels.

Main Methods:

• A semi-and-weakly supervised deep learning framework for image segmentation.
• Utilizing clinical tumor diameter measurements to guide pseudo-label generation.
• Employing accurate pseudo-labels to refine the segmentation model.

Main Results:

• The proposed method effectively leverages tumor diameter for weak supervision.
• Accurate candidate tumor regions were identified for pseudo-label selection.
• The generated pseudo-labels significantly improved segmentation performance.
• The method achieved superior performance compared to other comparative approaches.

Conclusions:

• Semi-and-weak supervision using clinical tumor diameter is effective for pathological image segmentation.
• The proposed method offers a practical solution for improving segmentation accuracy with limited annotations.
• This technique has the potential to enhance diagnostic capabilities in digital pathology.