FlowGAN Analysis Scripts

This folder contains the analysis scripts used to generate the results presented in the manuscript. The scripts analyze FlowGAN-generated synthetic FDG-PET images for two patient populations: Temporal Lobe Epilepsy (TLE) and Mild Cognitive Impairment (MCI).

Overview

The analysis pipeline compares three imaging modalities:

• Real FDG-PET: Ground truth metabolic imaging
• Synthetic FDG-PET: FlowGAN-generated images from ASL and T1w MRI
• ASL: Arterial Spin Labeling perfusion imaging

Two brain atlases are supported:

• DKT Atlas: Native space parcellation
• Harvard-Oxford Atlas: MNI space parcellation

Analysis Scripts

Script	Description
00_prepare_data.py	Data preparation from raw imaging files (requires original NIfTI data)
01_quality_metrics.py	Image quality metrics - SSIM, PSNR, RMSE, NCC (requires original NIfTI data)
02_regional_analysis.py	Regional correlation and Bland-Altman analyses
03_congruency_analysis.py	Sign congruency analysis for asymmetry direction
04_lateralization_cohens_d.py	Cohen's d effect sizes for clinical discrimination
manuscript_values.py	Extracts all manuscript values with statistics
utils.py	Shared utility functions
run_all.py	Master script to run all analyses

Note: Scripts 00_prepare_data.py and 01_quality_metrics.py require access to original NIfTI imaging data which are not included in this package. Reviewers should use the pre-computed pickle files and run analyses 02-04.

Data Files

Pickle Files (Pre-computed Regional Data)

File	Description
df_pet_merged.pkl	TLE dataset (DKT atlas)
df_pet_merged_mci.pkl	MCI dataset (DKT atlas)
df_pet_merged_ho.pkl	TLE dataset (Harvard-Oxford atlas)
df_pet_merged_mci_ho.pkl	MCI dataset (Harvard-Oxford atlas)

Each pickle file contains a pandas DataFrame with columns:

• subject: Subject identifier
• region_name: Brain region name
• side: Hemisphere (L/R)
• value_pet_original: Real FDG-PET SUVR
• value_pet_recon: Synthetic FDG-PET SUVR
• value_asl: ASL rCBF
• atlas_index: Atlas region index

Metadata Files (data/ folder)

File	Description
clinical_metadata.xlsx	TLE clinical metadata (lateralization info)
list_of_control_subjects.txt	MCI control subject list
list_of_MCI_subjects.txt	MCI patient subject list
dkt.csv	DKT atlas region definitions
ho.csv	Harvard-Oxford atlas region definitions

Running the Analyses

Prerequisites

Install required packages:

pip install -r requirements.txt

Run All Analyses

To run the complete pipeline (using pre-computed pickle files):

python run_all.py --skip-data-prep

The --skip-data-prep flag uses the pre-computed pickle files instead of regenerating them from raw imaging data.

Run Individual Scripts

Each script can be run independently:

python 02_regional_analysis.py
python 03_congruency_analysis.py
python 04_lateralization_cohens_d.py
python manuscript_values.py

Options

• --skip-data-prep: Use existing pickle files (recommended for reviewers)
• --tle-only: Run only TLE analyses
• --mci-only: Run only MCI analyses
• --include-ho: Include Harvard-Oxford atlas analyses

Output

Results are saved to:

• figures/: PDF and PNG figures organized by analysis
• tables/: CSV and Excel tables organized by analysis

FlowGAN Code

The flowgan_code/ directory contains the full FlowGAN pipeline for training and running inference. See flowgan_code/README.md for detailed instructions.

• Inference: Run FlowGAN on new ASL + T1 data to generate synthetic FDG-PET images using pretrained models (TLE or MCI cohort)
• Training: Train new FlowGAN models from scratch (standard, augmented, or k-fold cross-validation)

Quick start (inference):

cd flowgan_code/code
python run_FlowGAN.py --subs_file ../sample_list_of_subjects.txt --data ../sample_data/ --output_dir ../sample_data_outputs/ --model MCI