by John R. Fischer
, Senior Reporter | April 27, 2021
U.S. and Brazilian researchers have developed an AI-based solution that can automatically identify abnormal findings on brain MR scans.
Massachusetts General Hospital and Brigham and Women's Hospital Center for Clinical Data Science in Boston teamed up with Diagnosticos da America SA (DASA), a medical diagnostics company in Brazil, to create a triaging tool that could speed up care for those with various diseases and stroke or head-related injuries.
"The model has just been integrated in the radiology system of our collaborators at Dasa. The radiologist sees a new column that says 'normal' or 'abnormal' and it's possible to filter the studies based on this column. They are still experimenting with some small pilot studies to find out which specific workflow would be more beneficial," study co-lead author Dr. Romane Gauriau, former machine learning scientist at Mass General Hospital and Brigham and Women's Hospital Center for Clinical Data Science, told HCB News.
For those who need to move fast and expand clinical capabilities -- and would love new equipment -- the uCT 550 Advance offers a new fully configured 80-slice CT in up to 2 weeks with routine maintenance and parts and Software Upgrades for Life™ included.
The automated system utilizes a convolutional neural network to classify brain MR scans as either “likely normal” or “likely abnormal”. The model is the first of its kind to leverage a large and clinically relevant data set and use full volume MR data to detect overall brain abnormality.
A preliminary test on three large data sets of more than 9,000 examinations from different institutions on two continents showed the algorithm was able to distinguish likely normal from likely abnormal examinations. It was then validated by another test that used it on a data set acquired at a different time period and from a different institute than the one with the data used to train it.
It is expected to help identify incidental findings and be beneficial in outpatient care, according to the researchers. They next plan to evaluate its clinical utility and potential value for radiologists, as well as develop more classifications than just “likely normal” or “likely abnormal”.
Further evaluation is currently taking place in a controlled clinical environment in Brazil with research collaborators from DASA.
The findings were published in Radiology: Artificial Intelligence