Multi-Task Weakly-Supervised Attention Network for Dementia Status Estimation With Structural MRI

Abstract:

Noninvasive structural magnetic resonance imaging (MRI) can accurately predict neuropsychological test scores, which helps diagnose dementia (e.g., Alzheimer’s disease (AD)) and predict its progression. Existing machine/deep learning approaches preselect dementia-sensitive brain locations for MRI feature extraction and model construction, potentially causing stage heterogeneity and poor prediction performance. These methods also preselect brain locations using prior anatomical knowledge (e.g., brain atlas) and time-consuming nonlinear registration, ignoring individual-specific structural changes during dementia progression because all subjects share the same brain regions. We propose a multi-task weakly-supervised attention network (MWAN) for joint regression of multiple clinical scores from baseline MRI scans. MWAN has three sequential components: a backbone fully convolutional network for extracting MRI features; a weakly supervised dementia attention block for automatically identifying subject-specific discriminative brain locations; and an attention-aware multitask regression block for jointly predicting multiple clinical scores. MWAN is a fully trainable deep learning model that integrates dementia-aware holistic feature learning and multitask regression model construction. Our MWAN method estimated clinical scores of MMSE, CDRSB, and ADAS-Cog on two public AD data sets. Our method outperforms state-of-the-art regression methods in quantitative experiments. Qualitative results show that our MWAN method accurately identifies dementia-sensitive brain locations that retain individual specificities and are biologically meaningful.

Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.