At the end of June, Microsoft announced May-Dxo, an artificial intelligence capable, according to its tests, to diagnose four times better than human doctors, at a lower cost. This breakthrough is part of a broader dynamic: that of ultra-specialized medical AI, thoughts to support-and sometimes challenge-clinical gaze.
Advertisement
In this line, the Netherlands Cancer Institute and the Belgian company Robovision Healthcare reveal Brainmets.ai, a system capable of identifying brain metastases invisible to traditional MRI. His performances, already tested in real situations, arouse both enthusiasm … and the requirement of careful examination.
AI detects brain tumors invisible to MRI with 97 % precision
Up to 17 % of adults with cancer develop brain metastases. However, their early detection remains a challenge: some lesions are barely a few millimeters and are based in the noise of the dozens of MRI cuts.
It is on this ground that Brainmets.ai is essential. Trained on more than 1,500 MRI annotated by specialized radiologists, then independently tested on 311 clinical examinations, it displays a sensitivity of 97.4 % in terms of lesions. In other words, it identifies almost all of the anomalies identified by the experts-and even more.
The performance vary depending on the size of the lesions:
- 100 % detection for those measuring more than 12 mm,
- 98 % between 6 and 12 mm,
- 97.9 % between 3 and 6 mm,
- And another 93 % under 3 mm – where human detection becomes extremely complex.
[mediaid=”1200839″alt=”graphicshowingthesensitivityofbrainmetsatodetectbrainmetastasestoMRIaccordingtotheirsizefrom≥12mm(100%)to[mediaid=”1200839″alt=”GraphiquemontrantlasensibilitédeBrainMetsaipourdétecterlesmétastasescérébralesàl’IRMselonleurtaillede≥12 mm(100 %)à
Advertisement
