Development of explainable and reproducible artificial intelligence for medicine

Artificial intelligence (AI) holds the potential to revolutionize medicine and healthcare, especially in diagnosis and treatment. However, integrating AI into medicine presents several challenges that demand immediate consideration. This study examines three key aspects: explainability, reproducibility, and the scarcity of data due to privacy concerns. Explainability is vital for increasing trust in AI systems, especially in medical applications where decisions directly impact patient well-being. Reproducibility ensures the reliability of machine learning models across different settings. In this work, a new algorithm is proposed to compute average explanations to enhance these aspects. This approach aims to provide consistent and reproducible explanations, particularly in validation settings, contributing to the transparency and reliability of AI in medical decision-making. Additionally, privacy regulations intensify the scarcity of medical data, which prevents the development of effective AI models. In response, this investigation explores the potential of applying swarm learning (SL). Swarm learning is a recently proposed technology that empowers collaborative model training across decentralized data and computational sources while preserving data privacy. This innovative approach overcomes data scarcity issues and ensures compliance with stringent privacy regulations, preparing for a more robust AI development in the medical domain. This study underscores the necessity of addressing critical aspects such as explainability, reproducibility, and privacy concerns when deploying AI for healthcare applications.