Using machine learning, researchers have developed an algorithm that can accurately diagnose Alzheimer’s disease from a single MRI brain scan. The system is more accurate than any pre-existing diagnostic tool available to doctors and can also distinguish early-stage disease from more advanced stages.
Currently doctors have few tools available to easily diagnose Alzheimer’s disease. Alongside cognitive tests, spinal fluid can be tested for levels of certain toxic proteins associated with Alzheimer’s and some trained experts can detect neurodegeneration from brain scans. But there is a pressing need for a simple, more consistent way to diagnose this devastating disease. And here, researchers have turned to machine learning technology.
Using magnetic resonance imaging (MRI) data, the new research divided the brain into more than 100 distinct regions. An algorithm was then trained on a dataset of several hundred patients, some with Alzheimer’s, some with other neurological conditions, and some healthy controls.
Once trained the algorithm was put to the test on an independent set of brain scans and remarkably it was able to detect those patients with Alzheimer’s from healthy subjects with 98 percent accuracy. Even more impressively, the algorithm could differentiate early-stage Alzheimer’s brains from late-stage scans with 79 percent accuracy.
Interestingly, because the algorithm was looking at small Alzheimer’s-related changes across the entire brain, several regions were identified that had not been associated with Alzheimer’s in the past. Changes in regions such as the cerebellum and the ventral diencephalon were unexpected and the study calls for new research into the association between Alzheimer’s and these brain regions.
“Although neuroradiologists already interpret MRI scans to help diagnose Alzheimer’s, there are likely to be features of the scans that aren’t visible, even to specialists,” said Paresh Malhotra, a researcher from Imperial College London working on the project. “Using an algorithm able to select texture and subtle structural features in the brain that are affected by Alzheimer’s could really enhance the information we can gain from standard imaging techniques.”
Experts not affiliated with this new study are cautiously optimistic about the findings. Charles Marshall, a neurologist from Queen Mary University of London, says this new technology has the potential to revolutionize Alzheimer’s diagnoses but more real-world proof is needed before it is rolled out into clinics.
“For patients to benefit, we now need to evaluate how well machine learning technologies can detect Alzheimer’s disease in real-world clinical settings rather than using carefully curated research data,” Marshall said.
Rob Howard, from University College London, is a little more circumspect. He suggests the new findings are interesting but warns dementia diagnoses are profoundly unsettling for patients and inevitably will require more clinical work than just a single brain scan.
“A diagnosis of dementia is life-changing and should always be made after consideration of the patient’s history, the clinical examination and the results of tests such as brain scans,” noted Howard. “Over-reliance on brain imaging has been shown to be associated with dementia misdiagnoses and I have learned to be cautious with patients who tell me that their dementia was diagnosed from a brain scan.”
Lead on the new research, Eric Aboagye, said the new technology could be a tool that helps speed up a diagnostic process that currently can stretch out for months and months. The new technology also could be used to help researchers better identify early-stage Alzheimer’s patients for clinical trials testing new treatments.
“If we could cut down the amount of time they have to wait, make diagnosis a simpler process, and reduce some of the uncertainty, that would help a great deal,” said Aboagye. “Our new approach could also identify early-stage patients for clinical trials of new drug treatments or lifestyle changes, which is currently very hard to do.”
The new study was published in the journal Communications Medicine.
Source: Imperial College London