Evaluating AutoML Applications in Healthcare

This technical report discusses the challenges and opportunities associated with the adoption of automated machine learning (AutoML) in healthcare. It outlines the slower growth of AutoML in this sector compared to others, primarily due to the lack of transparency and the 'black-box' nature of many tools. The U.S. Food and Drug Administration (FDA) has initiated efforts to explore AutoML applications through the precisionFDA platform, encouraging innovators to assess its efficacy. The report details the approach taken by DRT Strategies in collaboration with CloudLeap Technologies to evaluate AutoML tools using brain cancer gene expression data. It describes the rigorous assessment of datasets, the selection of open-source AutoML tools, and the challenges faced in model training, particularly concerning biased datasets. Additionally, the report provides recommendations for improving data quality, democratizing AutoML tools, and ensuring human oversight in AI/ML applications. The findings highlight the potential benefits of AutoML in enhancing efficiency and accuracy in healthcare applications.