Code and instructions to run scripts for EE 559 (Sp2026) Final Project by Rama Chandra Bhogasamudram and Neil Patel.
- nibabel
- numpy, pandas, matplotlib
- pytorch
- pytorch-3dunet (conda install -c conda-forge pytorch-3dunet)
- Extract tissue volumes from pre-annotated segmentation masks → Ridge regression age prediction (baseline)
- Train 3 segmentation models on T2w images:
- Vanilla 3D CNN (encoder-only, small)
- Vanilla 3D CNN with Instance Norm
- 3D UNet Just to benchmark with popular and pre-defined architecture
- Loss:
CombinedLoss = ce_w·CrossEntropy + (1−ce_w)·SoftDiceLoss(background ignored) - Evaluation: per-class Dice score + mean Dice (foreground) + voxel accuracy (without background)
- End-to-End: Age Prediction from Predicted Segmentation of MRI
- Change folder path to dataset and participants.tsv