Introduction
RCR Training
Project Proposals
Title IX Training
Research Procedures
Team work Presentation
Read "Transformer Model for Alzheimer's Disease Progression Prediction using Longitudinal Data." By Mahdi Moghaddami, Clayon Schubring, Mohammad-Reza Siadat
Continued Literary Review by searching for Research papers about the topics of extracting fMRI Data into usable data and prediction models for Alzheimer's Disease
Began setting up the REU-2025 Project Website
Continued Reading Research Papers about GNNs and processing fMRI data.
Read "GNNs in neuroscience: graph convolutional networks for fMRI analysis"
Read "BrainGNN: Interpretable Brain Graph Neural Network for fMRI Analysis"
Skimmed through some other Research Papers
Created Week 1 Presentation
Meeting With Mahdi was Moved to Saturday Morning
Continued Literature Review
Read papers focused on multimodal models as well as how to solve the missing data problem
Found methods for imputation including KNN, RNN-based, soft-impute and Matrix Factorization
Some of the Key Papers I read were:
" Addressing the missing data challenge in multi-modal datasets for the diagnosis of Alzheimer's disease "
"Multimodal Data Fusion of Deep Learning and Dynamic Functional Connectivity Features to Predict Alzheimer’s Disease Progression "
"Application of MRI, fMRI and Cognitive Data for Alzheimer’s Disease detection"
Created Week 2 Presentation
Came up with some theories to be tested in the upcoming weeks
Developed a list of methods and models to be tested going forward and to narrow down which ones work the best with our goals
Learned how to use dcm2niix to convert our ADNI data from DICOM to NIFTI Format
Tried to use an adni-to-bids converter but had troubles getting the required files for it to properly run
Tested fMRIPrep and learned that if we used Free Surfer the time for processing would increase dramatically
Began downloading all patients data who have resting state fmri data and MPRAGE smri data for brain mapping
Learned the BIDS format and how to properly rename and restructure files and folders to allow for fMRIPrep to run
Began fMRI Prep preprocessing
Week 4: June 9 -June 13
fMRI Prep failed
Had to restart because all patients data corrupted when fMRI Prep Failed
Went through and removed all corrupted or incomplete patient visits that caused errors
Created scripts for finding missing or corrupted data and outputting it in a log
Found a way to utilize the full power of our A100 by running patients in parallel rather than one at a time bottlenecking the performance
Began looking code from research papers
Found GitHub from "STNAGNN: Data-driven Spatio-temporal Brain Connectivity beyond FC" for a Spatio-temporal GNN to be used for predictions based upon the fMRI data
Started testing imputation methods for the large amount of missing data in the patient visits from the tadpole dataset
Started looking into data and made graphs for midterm presentation based on patient diagnosis groups and visits length
Week 5: June 16 -June 20
Finished creating all FC Matrices along with labels
Began creating Dynamic FC Matrices and saving the results
Began working on STNAGNN Model and converting it to work with our data
Got STNAGNN working for baseline classification but didn't pursue increasing the base 60% accuracy with overfitting
Started trying to convert STNAGNN's model to work with prediction rather than exclusively classification
Week 6: June 23 -June 27
Continued Trying to work on fixing STNAGNN with little success
Couldn't get STNAGNN model to work with prediction with our dataset
Began working with the BrainGNN model
Attempted to make it work with Alzheimer's Disease Prediction
Trying to fix the overfitting problem with our dataset on the model
Began trying to learn how to work this into our own model going forward
Week 7: June 30 -July 4
Began working on our own model
Started looking at more literature papers for the purpose of finding Alzheimer's prediction related code and methods
Looked into "Auto-Metric Graph Neural Network Based on a Meta-Learning Strategy for the Diagnosis of Alzheimer’s Disease"
Began working on the GNN Layer
Based the GNN Encoder off of BrainGNN's model
Week 8: July 7 -July 11
Decided to scrap tabular idea for the scope of the program to focus more on the fMRI features
Now focusing on specifically if fMRI data can be useful when predicting Alzheimer's Disease Progression Prediction
Added Focal Loss to improve model for imbalanced dataset
Began working on poster
Began Running tests on Focal Loss alpha and gamma values to select best values
Began working on paper
Week 9: July 14 -July 18
Continued building and adding features to our model
Tested 3 different RNN based models with RNN, GRU, and LSTM
Also tested each model with a max visits variable limiting the max visits to test correlation
Worked on tweaking our model parameters to try and get the best results possible for our model
Implemented features to extract figures from our model for our poster
Add k-fold Cross Validation for final poster results
Submitted poster for printing
Continued working on research paper
Week 10: July 21 -July 25
Continued finishing up research report
Continued to clean the code base for submission
Worked on getting all deliverables ready and finished
Presented at MIDSURE 2025
Finished all REU Deliverables