A study published in Radiology showed that a deep learning model predicted Alzheimer’s disease with 82% specificity and 100% sensitivity about 6 years before diagnosis by using fluorine 18 fluorodeoxyglucose PET imaging studies of the brain.
"Differences in the pattern of glucose uptake in the brain are very subtle and diffuse," Jae Ho Sohn, MD, from the department of radiology & biomedical imaging at the University of California, San Francisco, said in a press release. "People are good at finding specific biomarkers of disease, but metabolic changes represent a more global and subtle process."
Researchers examined whether a deep learning algorithm could predict the final diagnosis of Alzheimer’s disease among patients who underwent fluorine 18 fluorodeoxyglucose PET of the brain.
The investigators collected prospective fluorine 18 fluorodeoxyglucose PET brain images from the Alzheimer’s Disease Neuroimaging Initiative dataset, which included 2,109 imaging studies from 1,002 patients, and a retrospective independent test set, which included 40 imaging studies from 40 patients. They trained the deep learning algorithm on 90% of the dataset, then tested it on the remaining 10% and the independent test set. The model was evaluated with sensitivity, specificity and receiver operating characteristic.
A deep learning model predicted Alzheimer’s disease with 82% specificity and 100% sensitivity about 6 years before diagnosis by using fluorine 18 fluorodeoxyglucose PET imaging studies of the brain, according to study findings.
The deep learning algorithm learned the metabolic patterns connected to Alzheimer's disease, according to the press release. The model achieved area under the receiver operating characteristic curve of 0.98 (95% CI, 0.94-1) when tested on its ability to predict the final Alzheimer’s disease diagnosis in the independent test set, Sohn and colleagues reported. The algorithm achieved 82% specificity at 100% sensitivity at detecting the disease an average of 75.8 months before final diagnosis.
After comparing the algorithm’s performance to that of radiologic readers’, the investigators also found the algorithm outperformed the readers (57% sensitivity and 91% specificity; P < .05).
"We were very pleased with the algorithm's performance. It was able to predict every single case that advanced to Alzheimer's disease,” Sohn said in the release. "If we diagnose Alzheimer's disease when all the symptoms have manifested, the brain volume loss is so significant that it's too late to intervene. If we can detect it earlier, that's an opportunity for investigators to potentially find better ways to slow down or even halt the disease process." – by Savannah Demko
Disclosures: Sohn reports grants from UCSF. Please see the full study for all other authors’ relevant financial disclosures.