Explaining COVID-19 diagnosis with Taylor decompositions

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• The COVID-19 pandemic necessitates advanced diagnostic tools.
• Computer-aided diagnosis using CT and X-ray images is crucial.
• Interpretability of AI models in medical diagnosis remains a challenge.

Purpose of the Study:

• To compare explainable artificial intelligence (XAI) approaches for COVID-19 diagnosis.
• To enhance the interpretability of deep learning models in medical imaging.
• To evaluate XAI methods using VGG11 and VGG16 deep networks.

Main Methods:

• Implemented and compared three XAI techniques: Composite Layer-wise Propagation, Single Taylor Decomposition, and Deep Taylor Decomposition.
• Utilized VGG11 and VGG16 deep neural networks for analysis.
• Assessed the explanation capabilities of XAI methods on COVID-19 diagnosis tasks.

Main Results:

• The study provides an insightful comparison of the selected XAI approaches.
• Each XAI method demonstrated unique strengths and weaknesses in explaining COVID-19 diagnoses.
• The findings highlight the potential of XAI in improving trust and understanding of AI-driven medical diagnostics.

Conclusions:

• Explainable AI offers promising avenues for enhancing the interpretability of AI-based COVID-19 diagnostic systems.
• Further research is needed to fully leverage XAI for robust and transparent medical image analysis.
• The comparative analysis serves as a foundation for future advancements in XAI for infectious disease diagnosis.