A new study finds that a single MRI could potentially be enough to accurately diagnose Alzheimer's disease. A team of researchers from the National Institute for Health and Care Research (NIHR) Imperial Biomedical Research Centre developed a predictive model that computes multigenerational, statistical, morpho-functional mesoscopic traits from T1-weighted MRI scans, with or without cognitive scores.
The machine learning technology evaluates structural features within the brain, including regions not previously associated with Alzheimer's disease, which investigators pointed out in Communications Medicine, is the most common cause of dementia (2022; https://doi.org/10.1038/s43856-022-00133-4), and causes "a progressive and irreversible deterioration of cognition that can sometimes be difficult to diagnose, leading to suboptimal patient care."
During the development of Alzheimer's disease, the brain shrinks "and the cells within it die. One method that can be used to assess brain function is magnetic resonance imaging," the authors wrote. For this study, the predictive method the researchers developed relied on MRI data to identify differences in the brain between people with and without Alzheimer's disease, including before clear shrinkage of the brain occurs, according to the investigators, who noted this method could ultimately be used to help diagnose patients with Alzheimer's disease.
Throughout the past 40 years, improved computational power and storage capacity have led to "numerous advances in developing non-invasive and low-cost structural biomarkers for Alzheimer's disease that combine neuroimaging approaches, in particular structural MRI, with machine learning," the authors wrote," noting that this approach involves the acquisition of image data, the segmentation of the region of interest, feature extraction, and selection for classification and prediction.
For each patient the researchers studied, a biomarker called Alzheimer's Predictive Vector (ApV) was derived using a two-stage least absolute shrinkage and selection operator (LASSO). Using data from the Alzheimer's Disease Neuroimaging Initiative, the investigators tested their method on brain scans of more than 400 patients with early- and later-stage Alzheimer's, healthy controls, and patients with other neurological conditions, such as frontotemporal dementia and Parkinson's disease, in addition to testing it with data from more than 80 patients undergoing diagnostic tests for Alzheimer's at Imperial College Healthcare NHS Trust.
According to the authors, the ApV reliably demonstrates between people with (ADrp) and without (nADrp) Alzheimer's-related pathologies-98 percent and 81 accuracy between ADrp, including the early form, mild cognitive impairment-and nADrp in internal and external hold-out test sets, respectively, without any prior assumptions or need for neuroradiology reads.
The new test is also superior to standard hippocampal atrophy (26% accuracy) and cerebrospinal fluid [beta]-amyloid measure (62% accuracy), according to the authors. A multiparametric analysis compared DTI-MRI derived fractional anisotropy, whose readout of neuronal loss agrees with ADrp phenotype, and SNPrs2075650 is "significantly altered in patients with ADrp-like phenotype."
In undertaking this study, the researchers hypothesized that the microstructure of Alzheimer's or mild cognitive impairment (MCI) would harbor subtle texture, size, and shape, and fractal features not visible to the naked eye, and that these features occurring at the mesoscopic scale could be detectable by machine learning, noted Eric Aboagye, PhD, Professor of Cancer Pharmacology and Molecular Imaging at Imperial College London and lead author of the study.
"We had done some work on ovarian cancer whereby we could assert prognosis by similarly training an algorithm to predict overall survival," said Aboagye, who is also Director of the CRUK-EPSRC-MRC-NIHR Comprehensive Cancer Imaging Centre.
For this study, Aboagye and his colleagues trained an algorithm by examining nearly 30,000 features from 115 brain regions and used machine learning to objectively pick a few features that would predict Alzheimer's or MCI.
In practicing the invention, one would take an MRI of the brain (T1 image), segment it automatically into the 115 regions and past use of the previously selected feature information together with reported weights to indicate whether or not a patient demonstrates Alzheimer's or MCI, he explained, noting that the algorithm selects features from regions that were known to be associated with Alzheimer's and those not normally associated with the disease.
While pointing out that further trails are needed, these findings could ultimately affect the way imaging and radiology teams approach the diagnosis and management of Alzheimer's, in that "the current qualitative approach to scoring MRI images, or CT, could benefit from highly accurate quantitative classification approaches," Aboagye concluded.
Mark McGraw is a contributing writer.