An algorithm is now able to diagnose Alzheimer’s from a single brain scan, according to a UK study. The modeling can even diagnose individuals at an early stage of the disease, normally a difficult task.
“Currently no other simple and widely available methods can predict Alzheimer’s disease with this level of accuracy, so our research is an important step forward,” said lead researcher Prof. Eric Aboagye in a media release.
The June 20 study published in Nature found that a machine-learning MRI algorithm can predict whether a person has Alzheimer’s or not with 98 percent accuracy. The modeling can also differentiate between an early and late-staged Alzheimer’s patient with a fair accuracy of 79 percent.
The modeling could be achieved on a standard 1.5 Tesla machine, which is commonly found in most hospitals.
Currently, to diagnose someone with Alzheimer’s, many tests must be taken, including MRI or CT, cognitive tests, blood tests, as well as tests to look for biomarkers or hallmarks of the disease.
However, all of the tests have limited accuracy by themselves, and need additional tests to support them. A 2017 study on biomarker data determined it had an accuracy of 77 percent for Alzheimer’s diagnosis, whilst a 2021 study found MRI could miss up to 20 percent of cases and may falsely predict at least 50 percent of cases for early diagnosis.
“Many patients who present with Alzheimer’s at memory clinics do also have other neurological conditions, but even within this group our system could pick out those patients who had Alzheimer’s from those who did not,” said Aboagye.
The modeling was made by adapting an algorithm used in classifying cancer tumors. The researchers divided the brain into 115 regions and allocated 660 different features such as size, shape, and texture. It is then trained to identify these features to accurately predict the presence of Alzheimer’s disease.
The researchers found the algorithm identified features that were previously not associated with Alzheimer’s such as the cerebellum—the part of the brain that maintains balance posture—and the ventral diencephalon which is linked to the sensory and motor functions and sleep-wake cycles.
These findings also open up potential new areas for research into these areas and their links to Alzheimer’s disease.
Since different features are present at different stages, and some people affected by Alzheimer’s may also have other diseases such as Parkinson’s, frontotemporal dementia, and so on; therefore the algorithm goes through two rounds using two different sets of criteria.
The two sets of criteria are called Alzheimer’s Predictive Vector 1 (ApV1) and Alzheimer’s Predictive Vector 2 (ApV2).
ApV1 is used in the first round to identify Alzheimer’s patients from those that are not affected. The algorithm examines 20 features across 14 regions across the brain out of 656 features. It also integrates cognitive scores and the presence of 19 Alzheimer’s disease hallmarks among 12 regions.
If an individual is identified with Alzheimer’s, they will then be scanned by the ApV2 algorithm which separates patients into early and late-stage disease. The ApV2 draws out 8 features and integrates cognitive scores with the presence of 19 Alzheimer’s disease hallmarks.
Consultant neurologist Dr. Paresh Malhotra said that this new algorithm can help to identify features that are not visible, even to specialists.
“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,” he said in the media release.
