AI Case Study
McGill University predicts signs of dementia two years before its onset with 84% accuracy with supervised learning
Scientists from the Douglas Mental Health University Institute’s Translational Neuroimaging Laboratory at McGill University in Canada used supervised machine learning to develop predictive models capable of recognising the early signs of dementia. Trained on PET scans of the brain the model is able to predict signs of dementia two years before its onset with 84% accuracy. This could help families and patients better prepare for treatment and care, and greatly improve clinical trial patient selection.
Public And Social Sector
Education And Academia
"The algorithm works by searching brain scans for the buildup of amyloid, a protein that accumulates in the brains of people who develop mild cognitive impairment and, later, dementia. Amyloid starts accumulating in the brain years, or sometimes decades, before the onset of dementia, and it does so at differnt rates and locations in the brain. What’s more, not everyone with amyloid buildup necessarily develops cognitive impairment. This has made spotting the development of dementia difficult for scientists.
Using hundreds of PET scans available through the Alzheimer’s Disease Neuroimaging Initiative (ADNI), scientists trained an algorithm to spot signs of dementia by having it first analyze the amyloid buildup in the scans of patients who had mild cognitive impairment. They then showed it brain scans taken before patients had developed the disease."
R And D
Core Research And Development
"Alzheimer’s disease is the most common form of dementia, accounting for approximately 60%-80% of the cases with alarming economic and social costs..."
"Scientists have long known that a protein known as amyloid accumulates in the brain of patients with mild cognitive impairment (MCI), a condition that often leads to dementia. Though the accumulation of amyloid begins decades before the symptoms of dementia occur, this protein couldn’t be used reliably as a predictive biomarker because not all MCI patients develop Alzheimer’s disease.
Imagine if doctors could determine, many years in advance, who is likely to develop dementia. Such prognostic capabilities would give patients and their families time to plan and manage treatment and care."
"..clinical trials could focus only on individuals with a higher likelihood of progressing to dementia within the time frame of the study. This will greatly reduce the cost and the time necessary to conduct these studies."
"The novel algorithm obtained an accuracy of 84% [to predict which patients would end up impaired] and an under-receiver operating characteristic curve of 0.91, outperforming the existing algorithms using the same biomarker measures and previous studies using multiple biomarker modalities. With its high accuracy, this algorithm has immediate applications for population enrichment in clinical trials designed to test disease-modifying therapies aiming to mitigate the progression to Alzheimer's disease dementia."
Logistic regression for feature selection. Random forest ensemble were used for prediction modelling.
"Using hundreds of PET scans available through the Alzheimer’s Disease Neuroimaging Initiative (ADNI), scientists trained an algorithm to spot signs of dementia by having it first analyze the amyloid buildup in the scans of patients who had mild cognitive impairment. They then showed it brain scans taken before patients had developed the disease."