Automatic prediction of cognitive and functional decline using baseline MRI and cognitive scores. (December 2021)

• Record Type: Journal Article
• Authors: Shafiee, Neda; Dadar, Mahsa; Ducharme, Simon; Collins, Louis
• Is Part Of: Alzheimer's & dementia. Volume 17(2021)Supplement 4
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.054217
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Patients in the early stages of AD dementia experience decline in their cognitive abilities at different rates. Accurately predicting the progression rate would enable the enrichment of patient populations in clinical trials. Using the early AD‐related pattern of atrophy and cognitive scores from a baseline visit, we trained a model to predict cognitive and functional decline in early stages of AD. Method: Data included 312 patients with mild Alzheimer's disease from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and were chosen based on amyloid positivity, the Clinical Dementia Rating (CDR=0.5) and the Mini‐Mental State Exam (MMSE, range 24‐30) scores. SNIPE (Scoring by Nonlocal Image Patch Estimator) was used to measure AD‐related atrophy patterns in the hippocampi (HC) and entorhinal cortex (EC) (Coupé et al. 2019). A balanced random forest was used to train models with feature sets including MR‐based z‐scored features (SNIPE scores for HC and EC), and cognitive test scores (Alzheimer's Disease Assessment Scale (ADAS‐13), MoCA (Montreal Cognitive Assessment), and MMSE) from baseline data. Classifiers were trained using different combinations of features plus age. The classification performance was evaluated based on the measured sensitivity, specificity, and accuracy to predict future functional and cognitive decline (defined as a 2‐point change in CSD‐SB). Result: Table 1 shows the classification performance of all trained models for two and three year follow up periods. Using hippocampal grading scores in addition to MoCA, ADAS‐13 and MMSE, yields the highest accuracy (76.9%) in predicting cognitive decline at 2 years. Figure 1 shows the ROC curve for two models: One consisting of all cognitive features and the other using both cognitive and MRI features. Comparing results between the classifier using only the baseline cognitive score and the corresponding classifier with the added MRI features showed that for both follow‐up periods, the accuracy of prediction is increased when adding MRI features. Conclusion: Microscopic changes due to AD pathology that occur in the HC and EC can be used as a feature in a predict the cognitive decline and increase the accuracy of prediction, even at the early stages of the AD trajectory.