Developing an explainable deep learning boundary correction method by incorporating cascaded x-Dim models to improve...

Area of Science:

• Medical Image Analysis
• Artificial Intelligence
• Explainable AI

Background:

• Deep learning excels in medical imaging but lacks transparency, hindering clinical adoption.
• Explainable AI (XAI) aims to increase trust by clarifying AI decision-making processes.
• Explainable segmentation methods are crucial for expert validation in clinical settings.

Purpose of the Study:

• To design an explainable deep correction method for medical image segmentation.
• To enhance the reliability and accuracy of AI-driven segmentation outputs.
• To improve the transparency and interpretability of deep learning models in medical analysis.

Main Methods:

• Incorporated cascaded 1D and 2D models for refining segmentation outputs.
• Implemented a 2-step iterative process involving local boundary validation and image patch segmentation.
• Applied a slice-by-slice refinement strategy to correct erroneous segmented regions.

Main Results:

• The proposed explainable method significantly improved segmentation accuracy over standard CNN models.
• Achieved state-of-the-art performance in 3D liver segmentation.
• Demonstrated an average Dice coefficient of 98.27% on the Sliver07 dataset.

Conclusions:

• The developed explainable deep correction method offers reliable and accurate segmentation results.
• This approach enhances user trust in AI-assisted medical image analysis.
• The method shows significant potential for clinical application in medical segmentation tasks.