Table 1

10 tips to get started with artificial intelligence healthcare projects

Phase 1: conceptualisation
1	Build a collaborative science team Recruit widely, select complementary expertise. Engage gatekeepers early.
2	Engage frequently with the end user What is the clinical problem to be addressed? End user engagement is a process, not an event. Develop digital transformation, not isolated algorithms.
3	Build collaboration agreements early Check for existing agreements. Use contracts and IG teams. Share your experiences with other CSTs.
4	Ethics: present a balanced view Present a balanced view of the challenges and benefits of AI projects. Seek out ethics review boards with prior experience. Ethical review may be protracted.
5	Invest in data science training for healthcare professionals in your team A common language is needed for the CST to work at its best.
Phase 2: data acquisition and preparation
6	Do not underestimate the challenge of data extraction Understand what data you really need. Identify a data champion. Engage healthcare informatics teams early; it will take longer than you expect. Consider building or contributing to a meta-data catalogue at your institution.
7	Protect patients’ data Give serious consideration to how you will anonymise and protect patient data from the outset of the project.
8	Remember that healthcare data varies in quality and reliability Healthcare data are influenced by a wide variety of factors. Clinicians can play a key role in interpreting and overcoming quality and reliability issues.
Phase 3: AI application
9	Design AI that can be trusted with patient care Use the right algorithms for your data and your end user. Strive for generalisable, well-validated solutions.
Phase 4: translation
10	Be mindful of medical device regulation Be mindful of the requirements for medical device regulation as your healthcare data science project progresses.

AI, artificial intelligence; CST, collaborative science team; IG, information governance.