Senior Design Projects

ECS193 A/B Winter & Spring 2019

Deep learning for Alzheimer’s diseases using EEG visibility graph

Email **********
Name Jiangyi Xia & John Olichney
Affiliation Center for Mind and Brain; Department of Neurology

Project's details

Project title Deep learning for Alzheimer’s diseases using EEG visibility graph
Background A major challenge of current research in Alzheimer’s disease ( AD ) is the development of sensitive and reliable ways to detect the disease in its earliest stages, when interventions and treatments are likely to be effective. In our lab, we study electrical brain activity during rest and tasks of attention, language, and memory using EEG. Our work shows that EEG measures are highly sensitive to early-stage AD and related cognitive impairments. Using the data we collected from patients and healthy elderly, we would like to build a deep learning framework for EEG-based classification of early stage AD.
Description EEG time series will be converted into Visibility Graph ( VG ), such that discriminative graph features can be extracted for classification analyses. VG preserves the structure and dynamics of the original EEG time series and is efficient in extracting information from noisy, multivariate, and nonstationary data such as EEG. VG features will be selected based on their sensitivity to AD. Deep neural networks ( CNN ) will be created using selected VG features to classify between different AD stages and between AD patients and healthy elderly. Multiple methods will be attempted to improve signal-to-noise ratio in the EEG data. These include averaging EEG trials during which specific stimuli and cognitive processes occur, i.e., cognitive ERPs.
Deliverable Software implementing VG feature abstraction and a DNN that take in multi-channel EEG time series and produce: ( 1 ) classification statistics, and ( 2 ) visualizations of VG features and EEG channels that contribute to accurate classification. The students will be expected to deliver the source code and provide comprehensive documentation of the code.
Skill set desirable Python ( or MATLAB ), TensorFlow ( or other deep learning frameworks )
Phone number **********
Client time availability 30-60 min weekly or more
IP requirement Open source project
Attachment Click here
Selected No
Team members N/A
TA N/A