AI is As Good As Humans to Interpret Diagnostics
[Thursday, September 26, 2019]

A study published in the British medical journal Lancet this week reported that Artificial Intelligence (AI) algorithms can read and analyze diagnostic imaging data at the same sensitivity and specificity as human professionals. Although the research was limited by the lack of studies directly comparing AI algorithms to human professionals, and the data used was from studies conducted between 2012 and 2019, the conclusions of the meta-analysis seems reasonable. The implications of the study are enormous. Medical imaging is the most common diagnostic tool used by healthcare professionals. However, reading medical images accurately and quickly is a challenge due to availability of trained human professionals and the cost of reading images that can be addressed by software based automated image analysis. The study should provide a good support to developer of AI algorithms for medical diagnostic applications. Although the Lancet report only reviewed AI used for medical imaging, the platform can be used for practically all kinds of diagnostic purposes. For FDA approval of AI for diagnostic, the developers should present data that can be demonstrated to address bias using studies preferably conducted in a clinical environment compared to computer simulations. Regulators currently expect systematic validation of the AI algorithms in clinical trials prior to approval and that requirement is likely not to change based on this new report but this should help build on the rationale and justification of the utility of these new software medical devices.