BA_fmri

BA_fmri is the current working repository for:

• brain-age model training
• guided backprop / saliency generation
• ROI generation from subject volumes
• ROI-level saliency summaries
• voxelwise group statistics

This README only records current positioning, canonical scripts, and the most common commands. It does not describe implementation details.

Current Canonical Scripts

• code/new_train_densenet169_BA_fmri.py Main training entrypoint. Produces checkpoint, age standardizer, validation predictions, and mean volume.
• code/guided_backprop_norm.py Main single-subject saliency generator.
• code/run_guided_bp_by_age_bins_norm.py Main batch saliency runner. It selects subjects and calls code/guided_backprop_norm.py.
• code/make_roi_from_npy.py Main ROI-generation entrypoint for single-subject or batch mode.
• code/compute_roi_saliency_density.py Main ROI saliency summary script.
• code/split_val_predictions_by_age_and_resid.py Main subject-selection CSV generator.
• code/compute_voxelwise_group_mwu.py Main voxelwise group-comparison script.

Most Common Commands

1. Train a model

python code/new_train_densenet169_BA_fmri.py \
  --db bhb_quasiraw_balanced_all \
  --epochs 40

2. Build a subject-selection CSV from validation predictions

python code/split_val_predictions_by_age_and_resid.py \
  --input_csv analysis/bhb_quasiraw_balanced_all/bhb_quasiraw_balanced_all_val_predictions_original.csv \
  --age_min 20 \
  --age_max 70 \
  --age_step 10 \
  --k_per_bin 3 \
  --mae_rank top

3. Run guided backprop in batch

python code/run_guided_bp_by_age_bins_norm.py \
  --dataset bhb_quasiraw_balanced_all \
  --age_min 20 \
  --age_max 70 \
  --age_step 10 \
  --k_per_bin 3 \
  --mae_rank top

4. Generate ROI folders in batch

python code/make_roi_from_npy.py \
  --split_input_csv analysis/bhb_quasiraw_balanced_all/bhb_quasiraw_balanced_all_val_predictions_original.csv \
  --age_min 20 \
  --age_max 70 \
  --age_step 10 \
  --k_per_bin 3 \
  --mae_rank top \
  --num_workers 4

5. Compute ROI saliency density in batch

python code/compute_roi_saliency_density.py \
  --split_input_csv analysis/bhb_quasiraw_balanced_all/bhb_quasiraw_balanced_all_val_predictions_original.csv \
  --age_min 20 \
  --age_max 70 \
  --age_step 10 \
  --k_per_bin 3 \
  --mae_rank top \
  --train_dataset bhb_quasiraw_balanced_all

More Documentation

• docs/pipelines.md
• docs/data_contracts.md
• docs/phase0_current_contracts.md