A new artificial intelligence-based algorithm can be applied to brain MRI scans to improve the diagnosis of Alzheimer’s disease in its early stages, UK researchers say.

Scientists at Imperial College London used machine learning technology to analyse structural features within the brain and identify differences between people with and without Alzheimer’s disease. They focused on morpho-functional mesoscopic traits from T1-weighted MRI scans, including before obvious neuronal loss and shrinkage of the brain occurred.

They developed a predictive model based on information from MRI images and cognitive scores from 781 people, including 208 with mild cognitive impairment due to Alzheimer’s, 181 with Alzheimer’s disease, 94 patients with Frontotemporal Dementia, and 84 with Parkinson’s disease, as well as 216 healthy controls.

Their “Alzheimer’s Predictive Vector” (ApV) reliably discriminated between people with and without Alzheimer’s-related pathologies.

On initial testing it had an accuracy of 98%  to discriminate between Alzheimer’s and non-Alzheimer’s disease brains. When validated on more than 480 patients undergoing diagnostic tests for suspected Alzheimer’s at Imperial College Healthcare NHS Trust, it had an overall accuracy of 81% in diagnosing Alzheimer’s disease, the researchers reported in Communications Medicine.

The new test was superior to standard hippocampal atrophy (26% accuracy) and cerebrospinal fluid beta amyloid measure (62% accuracy), they said. It was also able to distinguish between early- and late-stage Alzheimer’s with fairly high accuracy, in 79% of patients.

The researchers said their algorithm was based on the FreeSurfer software developed for use in classifying cancer tumours. They adapted it by dividing the brain into 115 regions and allocated 660 different features, such as size, shape and texture, to assess each region. They then trained the algorithm to identify where changes to these features could accurately predict the existence of Alzheimer’s disease.

They said the results showed that the MRI-based biomarker model had the potential to be incorporated into the clinical decision-making process for assessment of people with suspected Alzheimer’s disease such as those attending memory clinics.

“The [Alzheimer’s Predictive Vector] is reproducible and robust. It can be easily computed with the calculation of manually engineered features and is ready to be integrated into the clinical decision support system without the need for additional sampling or patient testing,” they wrote.

Professor Eric Aboagye, from Imperial’s Department of Surgery and Cancer, who led the research, said: “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. 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.

“Waiting for a diagnosis can be a horrible experience for patients and their families. 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. 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.”

He noted that the new system spotted changes in areas of the brain not previously associated with Alzheimer’s disease, including the cerebellum (the part of the brain that coordinates and regulates physical activity) and the ventral diencephalon (linked to the senses, sight and hearing). This opens up potential new avenues for research into these areas and their links to Alzheimer’s disease.

Dr Paresh Malhotra, who is a consultant neurologist at Imperial College Healthcare NHS Trust and a researcher in Imperial’s Department of Brain Sciences, said: “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. 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.”