2024 | Olga Ciobanu-Caraus, Anatol Aicher, Julius M. Kernbach, Luca Regli, Carlo Serra, Victor E. Staartjes
This review discusses the critical moment in machine learning (ML) in medicine, focusing on the challenges of reproducibility and interpretability. The exponential growth of ML publications in healthcare, driven by advancements in computational power and data availability, has led to a reproducibility crisis due to a lack of methodological rigor and standard reporting guidelines. Additionally, the complexity of modern ML models hinders their interpretability, which is crucial for clinical adoption. The review highlights the importance of reproducibility and interpretability, outlines the issues and challenges, and proposes solutions to address these problems. Key recommendations include developing standard reporting guidelines, encouraging data and code sharing, and using simpler models suitable for medical data. Model-agnostic explanation tools, sensitivity analysis, and hidden layer representations are also discussed as promising approaches to enhance interpretability. Balancing model performance and interpretability is essential for ensuring clinical applicability. The review emphasizes that addressing these issues is vital for the future evolution of ML in medicine.This review discusses the critical moment in machine learning (ML) in medicine, focusing on the challenges of reproducibility and interpretability. The exponential growth of ML publications in healthcare, driven by advancements in computational power and data availability, has led to a reproducibility crisis due to a lack of methodological rigor and standard reporting guidelines. Additionally, the complexity of modern ML models hinders their interpretability, which is crucial for clinical adoption. The review highlights the importance of reproducibility and interpretability, outlines the issues and challenges, and proposes solutions to address these problems. Key recommendations include developing standard reporting guidelines, encouraging data and code sharing, and using simpler models suitable for medical data. Model-agnostic explanation tools, sensitivity analysis, and hidden layer representations are also discussed as promising approaches to enhance interpretability. Balancing model performance and interpretability is essential for ensuring clinical applicability. The review emphasizes that addressing these issues is vital for the future evolution of ML in medicine.