MelaTools-ML systematic review

Are artificial intelligence/machine learning (AI/ML) algorithms ready for implementation in community and primary care settings to facilitate the early detection of skin cancer?

We systematically reviewed the academic literature to evaluate whether artificial intelligence/machine learning-based technologies are ready for implementation into primary and community care settings to help with the early diagnosis of skin cancer. We searched EMBASE, MEDLINE, Web of Science, and SCOPUS bibliographic databases identifying 11,296 studies, of which 272 were included in the final review.

Only 2 studies used data from low prevalence clinical settings to develop and test their algorithms, so we reviewed data from all 272 studies that could be relevant for primary care settings. Studies demonstrated reasonable diagnostic accuracy for melanoma (mean 89.5%, range 59.7-100%) and keratinocyte carcinomas (mean 86.7%, range 70.0-99.7%). However, we found large variability in study design, types of AI/ML techniques used, and evaluation methods. Many studies did not fully report their methods, making it hard to compare studies and to evaluate how relevant their findings are for primary care.

Overall, because of the methodological issues identified and the lack of studies set in primary care clinical settings, we concluded that widespread adoption into community and primary care practice cannot currently be recommended. We proposed a methodological checklist for use in development of new AI/ML algorithms to detect skin cancer, to facilitate their design, evaluation, and implementation.

Publication
Jones OT, Matin RN, van der Schaar M, Prathivadi Bhayankaram K, Ranmuthu CKI, Islam MS, Behiyat D, Boscott R, Calanzani N, Emery J, Williams HC, Walter FM. Artificial intelligence and machine learning algorithms for early detection of skin cancer in community and primary care settings: a systematic review. Lancet Digital Health. 2022. doi:10.1016/S2589-7500(22)00023-1