Alzheimer’s: Artificial intelligence predicts onset
The team responsible suggests that, after further validation, the tool could greatly assist the early detection of Alzheimer’s, giving treatments time to slow the disease more effectively.
The researchers, from the University of California in San Francisco, used positron-emission tomography (PET) images of 1,002 people’s brains to train the deep learning algorithm.
They used 90 percent of the images to teach the algorithm how to spot features of Alzheimer’s disease and the remaining 10 percent to verify its performance.
They then tested the algorithm on PET images of the brains of another 40 people. From these, the algorithm accurately predicted which individuals would receive a final diagnosis of Alzheimer’s. On average, the diagnosis came more than 6 years after the scans.
In a paper on the findings, which the Radiology journal has recently published, the team describes how the algorithm “achieved 82 percent specificity at 100 percent sensitivity, an average of 75.8 months prior to the final diagnosis.”
“We were very pleased,” says co-author Dr. Jae Ho Sohn, who works in the university’s radiology and biomedical imaging department, “with the algorithm’s performance.”
“It was able to predict every single case that advanced to Alzheimer’s disease,” he adds.
Alzheimer’s disease and PET imaging
The Alzheimer’s Association estimate that around 5.7 million people live with Alzheimer’s disease in the United States and that this figure is likely to rise to almost 14 million by 2050.
Earlier and more accurate diagnosis would not only benefit those affected, but it could also collectively save about $7.9 trillion in medical care and related costs over time.
As Alzheimer’s disease progresses, it changes how brain cells use glucose. This alteration in glucose metabolism shows up in a type of PET imaging that tracks the uptake of a radioactive form of glucose called 18F-fluorodeoxyglucose (FDG).
By giving instructions about what to look for, the scientists were able to train the deep learning algorithm to assess the FDG PET images for early signs of Alzheimer’s.
Deep learning ‘teaches itself’
The researchers taught the algorithm with the help of more than 2,109 FDG PET images of 1,002 individuals’ brains. They also used other data from the Alzheimer’s Disease Neuroimaging Initiative.
The algorithm utilized deep learning, a complex type of artificial intelligence that involves learning through examples, similarly to how humans learn.
Deep learning allows the algorithm to “teach itself” what to look for by spotting subtle differences among the thousands of images.
The algorithm was as good as, if not better than, human experts at analyzing the FDG PET images.
The authors note that “compared with radiology readers, the deep learning model performed better, with statistical significance, at recognizing patients who would go on to have a clinical diagnosis of [Alzheimer’s disease].”
Dr. Sohn cautions that the study was small and that the findings now need to undergo validation. This will involve using bigger datasets and more images taken over time from people at various clinics and institutions.
In the future, the algorithm could be a useful addition to the radiologist’s toolbox and improve opportunities for the early treatment of Alzheimer’s disease.
The researchers also plan to include other types of pattern recognition into the algorithm.
Change in glucose metabolism is not the only hallmark of Alzheimer’s, explains study co-author Youngho Seo, a professor in the Department of Radiology and Biomedical Imaging. Abnormal buildup of proteins also characterizes the disease, he adds.
“If FDG PET with [artificial intelligence] can predict Alzheimer’s disease this early, beta-amyloid plaque and tau protein PET imaging can possibly add another dimension of important predictive power.”
Prof. Youngho Seo
Source: Read Full Article