While AI may have a lot of potential to transform the future of medicine, humans are still much better at most complex intellectual tasks. The only thing that can be smarter than a human is a group of humans and new software from Unanimous AI, a company based in San Francisco, hopes to harness groups of doctors to improve the precision of diagnostic decisions.
The software, called Swarm AI, presented separate radiologists with chest X-rays of potential cases of pneumonia. The doctors answered questions regarding their findings and the results went into the software that supposedly uses some soft of artificial intelligence techniques to produce a result based on all the doctors’ findings.
The researchers that performed the study are presenting some findings at the 2018 Society for Imaging Informatics in Medicine’s Conference on Machine Intelligence in Medical Imaging (see abstract below). In summary, the average error rate using the Swarm AI system was 33% lower than diagnoses made by individual radiologists, while when comparing Swarm AI to CheXNet, a previously developed fully automated system from Stanford, the new software coupled with the doctors had a 22% higher accuracy rate.
Study abstract: Radiology SWARM: Novel Crowdsourcing Tool for CheXNet Algorithm Validation…
Unanimous AI company homepage…
Via: Unanimous AI…