Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

A New Algorithm for Shared Simultaneous Learning of Alzheimer's Disease Progression

Mutha, Pushkar (2021) A New Algorithm for Shared Simultaneous Learning of Alzheimer's Disease Progression. Master's Thesis, University of Pittsburgh. (Unpublished)

Primary Text

Download (1MB) | Preview


Alzheimer’s Disease (AD), a progressive neurodegenerative disease, is the most common form of dementia in older adults. It is preceded by stages of subtle cognitive decline called as Mild Cognitive Impairment (MCI), which is further stratified into Early (EMCI) and Late (LMCI) stages. Several imaging biomarkers are being investigated for early and accurate diagnosis as well as prognosis, and traditional approaches have generally focused on training multiple independent binary classifiers for distinguishing between Normal Controls (NC), EMCI, LMCI and AD subjects. However, these multiple one vs one classifiers could hold complementary information and sharing this information during the training may improve predictive performance.

We introduce a new framework to perform Shared Simultaneous Learning (SSL) of sparse logistic regression classifiers for NC vs EMCI, EMCI vs LMCI, and LMCI vs AD classification. We achieve this by adding a new term to the logistic loss function to enforce the weight vectors to be similar to each other. We introduce a constraint to minimize the squared Euclidean distance between the three weight vectors. A smooth approximation for the absolute value function is used and the model is optimized using gradient descent with line search. For each classifier, at the current gradient descent step, weights from the other two classifiers are shared.

We evaluated this algorithm on Structural Brain Connectome Networks generated from diffusion MRI of 202 subjects from the multicenter Alzheimer’s Disease Neuroimaging Initiative 2 (ADNI2) dataset. The normalized adjacency matrices were vectorized and passed as input for training along with the corresponding class labels. SSL outperformed independently trained multiple linear binary classifiers and achieved an average AUC of 0.53 for NC vs EMCI, 0.68 for EMCI vs LMCI, and 0.73 for LMCI vs AD classification. We also analyzed the brain connectivity patterns associated with highest odds ratio and show that abnormal inter-hemispheric connectivity patterns are indicative of EMCI vs LMCI whereas the right hemisphere of the brain is involved in the later stages.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Mutha, Pushkarpum6@pitt.edupum60000-0002-5961-7131
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Thesis AdvisorZhan,
Committee MemberAkcakaya,
Committee MemberDallal,
Date: 3 September 2021
Date Type: Publication
Defense Date: 13 July 2021
Approval Date: 3 September 2021
Submission Date: 20 July 2021
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 58
Institution: University of Pittsburgh
Schools and Programs: Swanson School of Engineering > Electrical and Computer Engineering
Degree: MS - Master of Science
Thesis Type: Master's Thesis
Refereed: Yes
Uncontrolled Keywords: Alzheimer's Disease, Logistic Regression, Machine Learning
Date Deposited: 03 Sep 2021 16:07
Last Modified: 03 Sep 2021 16:07


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item