Skip to main content
eScholarship
Open Access Publications from the University of California

Weakly Supervised Deep Feature Learning to Predict Aβ-positivity from Structural MR Brain Images

  • Author(s): Manuel, Cyrus
  • Advisor(s): Tosun-Turgut, Duygu
  • et al.
Abstract

Many novel treatments for Alzheimer’s Disease (AD) are aimed to target Aβ, one of the pathological hallmarks of AD, but are hampered by potential non responders due to lack of target Aβ pathology in their brains. Specifically, about ~25-40% of those clinically diagnosed with AD or mild cognitive impairment (MCI) would not have significant Aβ pathology. This study explores a deep learning framework to predict Aβ pathology positivity from baseline clinical assessments and structural MRI data routinely acquired from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Graph convolutional networks (GCNs) were trained on undirected graphs reconstructed from diffusion MRI and their performances was assessed to see their predictive value based on ground-truth Aβ-positivity estimates from AV45-PET scans. Anatomical brain parcellations with atrophy estimates from structural MRI constitute the vertices; tractography based connectivity estimates defined the edges of the graph model. A 10-fold cross validation on independent training and test sets were performed to assess the model performance in terms of classification accuracy, sensitivity, specificity, positive and negative predictive values. GCNs were able to learn from atrophy descriptors and network connectivity derived from MRI and predict Aβ-positivity. Atrophy was a significant predictor of Aβ-positivity in the AD model, but at a lesser degree in healthy and MCI models. The inclusion of other AD-related predictors showed: a significant improvement in test accuracy to 68±4%, sensitivity to 84±7%, specificity to 52±13%, negative predictive value to 77±5%, and positive predictive value to 64±4% in MCI models; and a significant improvement in test accuracy to 69±2% and specificity to 97±4% in HC models. Patterns of regional brain atrophy within large-scale brain networks might offer predictive value to whether or not a subject will test positive for an AV45-PET exam. Predictions are more accurate with the addition of well-established AD-related predictors, however more features may be necessary to increase the predictive ability in healthy and MCI subjects.

Main Content
Current View