PET Algorithm Allows Early Prediction of Alzheimer's

Deep learning model achieved 82% specificity, 100% sensitivity 75.8 months prediagnosis.
Deep learning model achieved 82% specificity, 100% sensitivity 75.8 months prediagnosis.

HealthDay News — Fluorine 18 (18F) fluorodeoxyglucose (FDG) positron emission tomography (PET) of the brain can be used to develop a deep learning algorithm for early prediction of Alzheimer's disease (AD) that has high specificity and sensitivity, according to a study published online November 6 in Radiology.

Yiming Ding, from the University of California in San Francisco, and colleagues collected prospective 18F-FDG PET brain images from the Alzheimer's Disease Neuroimaging Initiative (ADNI; 2109 imaging studies; 1002 patients) and a retrospective independent test set (40 imaging studies; 40 patients). A convolutional neural network of InceptionV3 architecture was trained on 90% of the ADNI data set and was tested on the remaining 10% and the independent test set. 

Related Articles

The researchers found that when evaluated on predicting the final clinical diagnosis of AD in the independent test set (82% specificity at 100% sensitivity), the algorithm achieved an area under the receiver operating characteristic (ROC) curve of 0.98 an average of 75.8 months before the final diagnosis. In ROC space, the algorithm outperformed reader performance (57% sensitivity, 91% specificity). Attention to known areas of interest was demonstrated on a saliency map, with a focus on the entire brain.

"With further large-scale external validation on multi-institutional data and model calibration, the algorithm may be integrated into clinical workflow and serve as an important decision support tool to aid radiology readers and clinicians with early prediction of AD from 18F-FDG PET imaging studies," the authors write.

Several authors disclosed financial ties to the pharmaceutical and/or technology industries.

Abstract/Full Text