Objective: A deep learning-based classification system (DLCS) which uses structural brain magnetic resonance imaging (MRI) to diagnose Alzheimer's disease (AD) was developed in a previous recent study. Here, we evaluate its performance by conducting a single-center, case-control clinical trial.
Methods: We retrospectively collected T1-weighted brain MRI scans of subjects who had an accompanying measure of amyloid-beta (Aβ) positivity based on a 18F-florbetaben positron emission tomography scan. The dataset included 188 Aβ-positive patients with mild cognitive impairment or dementia due to AD, and 162 Aβ-negative controls with normal cognition. We calculated the sensitivity, specificity, positive predictive value, negative predictive value, and area under the receiver operating characteristic curve (AUC) of the DLCS in the classification of Aβ-positive AD patients from Aβ-negative controls.
Results: The DLCS showed excellent performance, with sensitivity, specificity, positive predictive value, negative predictive value, and AUC of 85.6% (95% confidence interval [CI], 79.8-90.0), 90.1% (95% CI, 84.5-94.2), 91.0% (95% CI, 86.3-94.1), 84.4% (95% CI, 79.2-88.5), and 0.937 (95% CI, 0.911-0.963), respectively.
Conclusion: The DLCS shows promise in clinical settings where it could be routinely applied to MRI scans regardless of original scan purpose to improve the early detection of AD.
Keywords: Alzheimer disease; Clinical trial; Deep learning; Magnetic resonance imaging.