by John R. Fischer
, Senior Reporter | March 10, 2023
Researchers at Massachusetts General Hospital have developed a new method for detecting Alzheimer’s disease with AI and medical imaging, and even tested it in real-world clinical settings.
Using deep learning, the researchers extracted large amounts of data from routinely collected high-quality brain MR images from Alzheimer’s and non-Alzheimer’s patients, using the information to train AI models.
They then tested the model, which they named MUCRAN (Multi-Confound Regression Adversarial Network), across five data sets to evaluate if it could accurately identify Alzheimer’s disease based on real-world clinical data, regardless of which hospital the information came from and the time when the scans were performed.
Ad Statistics Times Displayed: 98874
Times Visited: 1327
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 research involved 11,103 images from 2,348 patients at risk for Alzheimer’s disease, and 26,892 images from 8,456 patients without Alzheimer’s disease. The model detected risk for the disease with 90.2% accuracy across all data sets.
“In real-world settings, clinical data in many hospitals, such as Mass General Hospital, Brigham Women's Hospital, and other satellite clinics, are very heterogeneous, with many confounding factors that could affect and confuse AI solutions when diagnosing. Our approach was to develop an AI solution that detects and regresses these confounding factors during training, such that the decision relies more on valid biomarkers rather than confounding factors,” Hyungsoon Im, assistant professor of MGH and Harvard Medical School, told HCB News.
Additionally, Im and his colleagues incorporated an uncertainty metric for the model to determine if patient data was too different from what it was trained on to make a successful prediction.
This, he says, is due to the fact that technical and demographical variances exist among hospitals’ and foreign countries’ data, equipment, and patient populations. In other words, the clinical data from one hospital will not necessarily generalize to data from another hospital.
“We are also developing AI solutions that take multiple parameters and types of data, such as other neuroimaging data (e.g., PET images), neurological test results, changes from previous examinations, and more, which would better mimic current clinical practice by clinicians when they diagnose AD,” said Im.
Data sets included images from MGH before 2019; Brigham and Women’s Hospital pre- and post-2019; and outside systems pre- and post-2019.
The findings were published in PLOS ONE
. Back to HCB News