I found what looks to be the same system on Pubmed[1] because the linked article lacked any measures of sens/spec. I do like how they ran it in two different workflows, of which the high sensitivity model seems most practical. Ideally if the high sensitivity model can screen out the negative cases it would free up resources for radiology to then parse through the ones flagged by AI, with the expectation being that the AI will throw a higher number of false positives that the radiologists can filter out. Overall case load ideally would decrease, especially as the model continues to improve.<p>1) <a href="https://pubmed.ncbi.nlm.nih.gov/37987665/" rel="nofollow">https://pubmed.ncbi.nlm.nih.gov/37987665/</a>