Although investigators have made strides in detecting signs of Alzheimer’s disease using high-quality brain imaging tests collected through research studies, a team at Massachusetts General Hospital (MGH) recently developed an accurate method of detection based on routinely collected clinical images of the brain. Advances could lead to more accurate diagnoses.
For the study published in PLUS ONEMatthew Leming, PhD, a research fellow at the MGH Center for Systems Biology and a researcher at Massachusetts Alzheimer’s Disease Research Center, and his colleagues used deep learning — a type of machine learning and artificial intelligence that uses large and complex data sets to train models.
In this case, scientists developed a model for detecting Alzheimer’s disease based on data from brain magnetic resonance images (MRIs) collected from patients with and without Alzheimer’s disease studied at the MGH before 2019.
Next, the group tested the model on five data sets — MGH after 2019, Brigham and Women’s Hospital before and after 2019, and external systems before and after 2019 — to see if it could accurately detect Alzheimer’s disease based on real clinical data, independent of hospital and time.
Overall, the research included 11,103 images from 2,348 patients at risk of Alzheimer’s and 26,892 images from 8,456 patients without Alzheimer’s disease. Across all five datasets, the model detected Alzheimer’s risk with an accuracy of 90.2%.
One of the key innovations of the work was the ability to detect Alzheimer’s disease independently of other variables such as age. “Alzheimer’s disease typically occurs in older adults, and as a result, deep learning models often struggle to detect the rarer, early-onset cases,” says Leming. “We addressed this by ‘blinding’ the deep learning model to features of the brain that it found to be overly associated with the patient’s reported age.”
Leming notes that another common challenge in disease detection, especially in real-world settings, is dealing with data that is vastly different from the training dataset. For example, a deep learning model trained with MRIs from a scanner manufactured by General Electric cannot recognize MRIs acquired with a scanner manufactured by Siemens.
The model used an uncertainty metric to determine whether patient data differed too much from what it was trained on to make a successful prediction.
“This is one of the few studies that has routinely used brain MRIs collected to try to detect dementia. While a large number of deep learning studies have been conducted on detecting Alzheimer’s from brain MRIs, this study has taken significant steps to actually do so in the real world. global clinical settings as opposed to perfect laboratory settings,” said Leming. “Our results – with generalizability across sites, across time and across populations – make a strong case for the clinical use of this diagnostic technology.”
Other co-authors are Sudeshna Das, PhD and Hyungsoon Im, PhD.
This work was supported by the National Institutes of Health and the Technology Innovation Program funded by the Ministry of Commerce, Industry and Energy of the Republic of Korea and managed through a subcontract with MGH.