Accessible artificial intelligence
In order for new technology to fulfill its potential in healthcare, clinicians must take an active role in its development and validation. Broadening access to AI tools and improving AI literacy is essential.
Automated machine learning
Automated machine learning (autoML) allows non-specialists to develop and deploy AI tools for a wide variety of tasks, including medical image analysis. In a systematic review, we explored all existing clinical applications of autoML, and identified strengths and weaknesses of different teams' approaches and reporting. Some 26 autoML platforms with a variety of technical requirements have been used for clinical work, but studies tend to be retrospective, limiting their implications for actual clinical practice. However, autoML often compares favourably to conventional machine learning and non-machine learning techniques, lowering the bar to entry for development of useful clinical applications. For more details, see our report in Annals.
To help clinicians interesting in using autoML in their work, we produced a tutorial which can act as a practical guide to development and validation; published in JMIR.
Educating future clinicians
To explore what clinicians should learn about artificial intelligence, we looked at how we envisage doctors and allied healthcare professionals interacting with AI in the future. Most clinicians are likely to be 'consumers' using AI tools without participating actively in development or validation. Clinicians with greater interest may become 'developers' helping to build new tools, or 'translators' helping validate and implement tools that can benefit patients and practitioners.
AI is predicted to significantly impact healthcare, so we looked to evidence-based medicine (EBM) as a previous paradigm shift that has required teaching at medical school to adapt. The progress and pitfalls of EBM teaching help inform a roadmap for implementation of AI teaching for medical students; further development of EBM teaching is also essential to ensure that new applications are tested rigorously to ensure that patients remain safe and receive optimal care. Our vision, published in Cell Reports Medicine, can inform curriculum designers to help best prepare clinicians for the future.
