Alibaba Open-Sources Medical AI That Achieves AUC 0.913 On 40,000 Scans

Recently, Alibaba announced they open sourced their medical AI model, Damo Radar. The model is capable of analyzing contrast enhanced CT scans and diagnosing over 150 kinds of abdominal diseases including cancer. According to a new research published in Science, the model has been tested on 40,000 real world cases. The model achieved an average AUC score of 0.913 (where 1.0 is perfect) for 146 kinds of clinical findings, outperforming most radiologists.

Of course, the model is not meant to replace doctors. That’s obvious. But having access to an open source medical AI model like this could be a good starting point. Now we don’t need to build everything from scratch to match the accuracy of many experts in detecting cancer.