An Explainable AI Algorithm for COVID 19 Detection Based on Layer wise Relevance Propagation скачать в хорошем качестве

An Explainable AI Algorithm for COVID 19 Detection Based on Layer wise Relevance Propagation

250408 Department of Computer Science & Engineering An Explainable AI Algorithm for COVID-19 Detection Based on Layer-wise Relevance Propagation An Explainable AI Algorithm for COVID-19 Detection Based on Layer-wise Relevance Propagation Abstract: We employ Explainable AI (XAI) to investigate the decision-making processes of a Convolutional Neural Network (CNN) built using TensorFlow, developed for classifying lung X-ray images into healthy lungs (normal), Pneumonia, and COVID-19 categories. Utilizing the XAI library, Innvestigate, while integrating Layer-Wise Relevance Propagation (LRP) and Deep Taylor Decomposition, we generate interpretive heatmaps to reveal critical regions influencing the model’s predictions. A comparative analysis of multiple LRP Rules including Alpha2, Beta1, and Z Plus, highlight differences in interpretability and precision, with methods such as LRP Z Plus and Deep Taylor Bounded offering enhanced contrast and clearer visualizations. These insights provide a detailed understanding of the model’s decision pathways, enabling the identification of potential biases and inaccuracies. Our findings not only demonstrate the utility of XAI in improving transparency and reliability in medical imaging AI, but also establish a framework for broader applications in Machine Learning where interpretability is paramount. RESEARCHERS: Vedant Hathalia, Bellarmine College Preparatory ‘27; Tolulope Elegbede, James Logan High School ‘26; Krithi Tandyala, James Logan High School ‘26; Josh Karthikeyan, American High School ‘27 ADVISOR: Viktoriia Liu Lab, Chemistry & Computer Neurobiology & Explainable AI & Augmented Reality KEYWORDS: Explainable AI (XAI) | Medical Imaging | Convolutional Neural Networks (CNN) | Layer-Wise Relevance Propagation (LRP) | Model Interpretability