IMPA-Net: Interpretable Multi-Part Attention Network for Trustworthy Brain Tumor Classification from MRI

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computational Neuroscience

Background:

• Deep learning (DL) models excel in medical image analysis, particularly brain tumor classification.
• The
• black box
• nature of DL hinders trust and clinical adoption due to opaque reasoning.

Purpose of the Study:

• To develop an interpretable multi-part attention network (IMPA-Net) for brain tumor classification.
• To enhance the interpretability and trustworthiness of DL-based classification outcomes in neuro-oncology.

Main Methods:

• Developed IMPA-Net, a novel interpretable deep learning architecture.
• Integrated global and local explanation mechanisms for model transparency.
• Utilized the BraTS2017 dataset for model training and validation.

Main Results:

• IMPA-Net achieved 92.12% classification accuracy for brain tumors.
• 86% of learned feature patterns were validated by radiologists as medically relevant.
• 81.17% of predictions were deemed trustworthy based on local explanations.

Conclusions:

• IMPA-Net offers a verifiable and trustworthy approach to glioma classification.
• The model's interpretability facilitates clinical decision support for health workers and patients.
• This interpretable DL model addresses the limitations of "black box" approaches in medical diagnostics.