Although researchers have made significant progress in using high-quality brain imaging (research level) tests to detect signs of Alzheimer’s disease, a team at Massachusetts General Hospital (MGH) recently developed an accurate detection method that relies on routinely collected clinical brain images. This progress may lead to more accurate diagnosis.
In this study, published in the journal PLOS ONE, Dr. Matthew Leming, a researcher at the Massachusetts General Hospital Center for Systems Biology and the Massachusetts Alzheimer’s Disease Research Center, and his colleagues used deep learning, a machine learning that uses large amounts of data and complex algorithms to train models, and artificial intelligence to detect Alzheimer’s disease.
The study was published in the journal PLOS ONE (latest impact factor: 3.752) on March 2, 2023
In this context, scientists have developed an Alzheimer’s disease detection model based on brain magnetic resonance imaging data of patients with and without Alzheimer’s disease seen at MGH before 2019.
Next, the team tested the model on five datasets – MGH after 2019, Brigham and Women’s Hospital before and after 2019, and external medical systems before and after 2019 – to see if it can accurately detect Alzheimer’s disease based on real-world clinical data, regardless of which hospital and time.
Overall, this study involved 11103 images from 2348 Alzheimer’s disease risk patients and 26892 images from 8456 non Alzheimer’s disease patients. In all five data sets, the accuracy of the model in detecting Alzheimer’s disease risk was 90.2%.
The main innovation of this work is that it can detect Alzheimer’s disease without considering other variables such as age. “Alzheimer’s disease usually occurs in elderly people, so deep learning models are often difficult to detect rare early onset cases,” Leming said. “When it is found that brain features are excessively correlated with the patient’s age, we solve this problem by making deep learning models’ blind ‘to these brain features.”
Leming pointed out that another common challenge in disease detection, especially in real-world environments, is processing data that is very different from training sets. For example, deep learning models trained on GE manufactured scanners may not be able to recognize the magnetic resonance images collected on Siemens manufactured scanners.
The model uses uncertainty measures to determine whether patient data is too different from training data to successfully predict.
“This is the only study that attempts to detect dementia using routinely collected brain magnetic resonance imaging. Although a large number of in-depth learning studies have been conducted to detect Alzheimer’s disease using brain magnetic resonance imaging, this study has taken a substantial step towards practical implementation in a real-world clinical environment rather than a perfect laboratory environment,” Leming said, “Our findings are universal across locations, times, and populations, providing strong evidence for the clinical application of this diagnostic technique.”