Extending Grad-CAM to DualNet.

Area of Science:

• Medical Imaging
• Machine Learning
• Computer Vision

Background:

• DualNet is a machine learning architecture designed for disease classification using frontal and lateral X-ray images.
• Current limitations of DualNet include a lack of integrated explainability methods, hindering clinical trust and adoption.
• Explainable AI (XAI) is crucial for understanding and validating the decisions made by complex machine learning models in healthcare.

Purpose of the Study:

• To extend the pytorch-grad-cam package for application with the DualNet architecture.
• To demonstrate the technical feasibility of applying explainability methods to DualNet for X-ray analysis.
• To lay the groundwork for future explainable AI research in DualNet-based medical diagnostics.

Main Methods:

• Adaptation of the pytorch-grad-cam library to interface with the DualNet machine learning model.
• Implementation of gradient-based class activation mapping techniques.
• Validation in an artificial scenario to confirm technical functionality.

Main Results:

• Successful extension of pytorch-grad-cam for use with DualNet.
• Demonstration of technical proof-of-principle for explainability in an artificial setting.
• The implemented methods provide visual insights into DualNet's decision-making process.

Conclusions:

• The adapted pytorch-grad-cam package is technically capable of providing explainability for DualNet.
• This work represents a foundational step towards enhancing the transparency of DualNet in medical image analysis.
• Further research is warranted to validate these explainability methods in real-world clinical scenarios.