Interpretable Model Based on Pyramid Scene Parsing Features for Brain Tumor MRI Image Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Deep learning models, particularly convolutional neural networks (CNNs), are often "black boxes," limiting interpretability in medical applications.
• Lack of interpretability hinders trust and adoption of computer-aided diagnosis (CADx) systems in clinical practice, especially for brain tumor segmentation.

Purpose of the Study:

• To develop an interpretable deep learning framework for brain tumor segmentation on magnetic resonance images (MRIs).
• To enhance the trustworthiness and clinical applicability of deep learning-based medical image analysis.

Main Methods:

• Proposed an interpretable deep learning image segmentation framework utilizing a pyramid structure.
• Integrated a gradient-based class activation mapping (CAM) method to visualize and explain the model's decision-making process.
• Trained and validated the framework on the public BraTS2018 dataset.

Main Results:

• The pyramid structure effectively captured global context, improving brain tumor segmentation performance.
• Class activation mapping successfully visualized feature importance across different layers of the pyramid structure, enabling model interpretation.
• Analysis of visualization results led to identified shortcomings and subsequent improvements in the pyramid model architecture.

Conclusions:

• The developed interpretable method effectively explains the role of the pyramid structure in brain tumor segmentation.
• The approach offers a valuable strategy for applying interpretable AI in medical image analysis.
• Demonstrates practical value for evaluating and optimizing brain tumor segmentation models.