Early Prediction of Alzheimer's using Dynamic Functional Connectivity and Deep learning
Poster No:
322
Submission Type:
Abstract Submission
Authors:
YUXIANG WEI1, Anees Abrol2, Vince Calhoun3
Institutions:
1Georgia Institute of Technology, Atlanta, GA, 2Georgia State University, Atlanta, GA, 3GSU/GATech/Emory, Decatur, GA
First Author:
Co-Author(s):
Introduction:
Alzheimer's disease (AD) is a neurodegenerative brain disorder that gradually transitions from asymptomatic pathological changes to clinical symptoms. Early diagnosis is pivotal in implementing proper treatment and potentially slowing disease progression. Functional magnetic resonance imaging (fMRI) has emerged as an non-evasive method capable of accurately capturing brain activities. In particular, the fMRI features estimated by dynamic functional connectivity approach models the dynamism of brain function, position itself as a promising biomarker for identifying AD and mild cognitive impairment symptoms. Nevertheless, studies probing asymptomatic at-risk subjects using fMRI remain relatively limited. The recent advance of deep learning enhanced the efficacy of encoding high-level information from brain dynamism, signifying a promising avenue for pre-symptom AD detection and analysis.
Methods:
In this work, we introduce a transformer-convolution-based framework, building on our previous work, for predicting and analyzing subjects that are at risk for AD. We propose an innovative spatial-temporal self-attention module to learn both the spatial dependencies across brain networks and temporal contextual variations. We validate our method based on the Emory Healthy Brain Study dataset and study 303 cognitive normal and 59 high-risk subjects.
Results:
Compared to other standard machine learning methods such as logistic regression that has 78.13% accuracy but 13.48% f1 score and 10.17% sensitivity on the high-risk subjects, the proposed method achieves 76.76% accuracy, 45.09% f1 and 66.10% sensitivity. To further study which brain network contributes to the final prediction, we provide interpretable analysis over the proposed framework based on the gradient-based interpretable method and present the saliency map.
Conclusions:
As such, the proposed method reveals distinct relations between various brain networks and AD progression, offering a promising direction for the study of asymptomatic AD with fMRI.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Data analysis
DISORDERS
FUNCTIONAL MRI
Machine Learning
MRI
Nerves
1|2Indicates the priority used for review
Provide references using author date format
Wierenga, C.E., Bondi, M.W. (2007) 'Use of Functional Magnetic Resonance Imaging in the Early Identification of Alzheimer's Disease'. Neuropsychol Rev 17, pp: 127–143