Dear all - what’s our thoughts on a new Chat GPT4 Medical Reporting Model being trialed over in the States? Preliminary findings seem to be that it is more accurate than human reporters (96% concordance when double reported vs. 92% concordance for human v human reporting on both plain film (all types) on the bulk of routine studies (simple to complex CT studies).
A point which surprised me was that the algorithm also follows template reporting and clinical coding rules precisely, with both patients and the US facility administrators (money people!) feeding back higher satisfaction on readability, tone and grammar. As many PACS are now cloud based, it seems to be a simple bolt-on for PACS vendors and pretty much slashes reporting backlogs to zero over the course of a few hours.
Obviously the UK has a very different way of working and treating clinical staff, but with Google including GPT-3 in search, Microsoft including it in the NHS .net Office 365 tenant and news GPT-4 is to be included “for free” over here in a PACS Based Reporting module this month should we be more concerned or take more steps to vet and assess this ‘march of the machines’ in advance?
Al.