A team at Massachusetts General Hospital (MGH) recently created an accurate method for detection that depends on routinely collected clinical brain images. Despite the fact that researchers have made progress in spotting signs of Alzheimer’s disease using high-quality brain imaging tests gathered as part of research studies. The development might result in a more precise diagnosis.
How will AI help in Alzheimer’s disease detection?
Matthew Leming, Ph.D., a research fellow at MGH’s Center for Systems Biology and an investigator at the Massachusetts Alzheimer’s Disease Research Center, and his associates used deep learning. It is a branch of machine learning and artificial intelligence that employs substantial amounts of data and intricate algorithms to train models—for the study, which was published in PLOS ONE.
Based on information from brain magnetic resonance imaging (MRIs) taken from patients with and without Alzheimer’s disease who visited MGH prior to 2019, the researchers in this case created a model for Alzheimer’s disease identification.
After that, the team evaluated the model’s performance using five different datasets, including MGH post-2019, Brigham and Women’s Hospital pre- and post-2019, and outside systems pre- and post-2019. In addition, it is to see if it could reliably identify Alzheimer’s disease based on clinical data collected in the real world, regardless of the location or time.
11,103 photos from 2,348 people at risk for Alzheimer’s disease and 26,892 photographs from 8,456 patients without the condition were used in total for the study. The program correctly identified the risk of Alzheimer’s disease in 90.2% of the five datasets.
The work’s capacity to identify Alzheimer’s disease regardless of other factors, such as age, was one of its primary advances. According to Leming, since Alzheimer’s disease commonly affects older persons, deep-learning models frequently have trouble identifying the more uncommon early-onset instances. In order to combat this, “we made the deep learning algorithm ‘blind’ to brain traits that it finds to be excessively correlated with the patient’s claimed age,” the researchers wrote.
Leming points out that working with data that varies significantly from the training set is another frequent difficulty in disease identification, particularly in real-world contexts. For instance, a deep learning model trained on MRIs from a General Electric scanner might not be able to distinguish between MRIs gathered on a Siemens scanner.
In order to establish whether the patient data were too dissimilar from those it had been trained on for it to be able to make a reliable forecast, the model employed an uncertainty metric.
“This is one of the few studies that tried to identify dementia using routinely gathered brain MRIs. Although there have been other deep learning studies for detecting Alzheimer’s using brain MRIs, this study took a significant step closer to carrying it out in actual clinical situations as opposed to ideal laboratory conditions, according to Leming. “Our results provide significant support for clinical application of this diagnostic method with cross-site, cross-time, and cross-population generalizability.”
Sudeshna Das, Ph.D., and Hyungsoon Im, Ph.D. are also co-authors.
MGH undertook this project under a subcontract with the National Institutes of Health and the Ministry of Commerce, Industry, and Energy of Korea.
Massachusetts General Hospital information
The original and biggest teaching hospital of Harvard Medical School is Massachusetts General Hospital, which was established in 1811. With more than $1 billion in annual research operations and more than 9,500 researchers spread across more than 30 institutes, centers, and departments, the Mass General Research Institute runs the biggest hospital-based research program in the country.
