/MRIAnatEval

Primary LanguagePython

MRIAnatEval

Git repository for MICCAI 2024

Metrics included:

  1. FID
  2. MMD
  3. MS-SSIM
  4. PCA-tSNE
  5. Segmentation Quality Control

Models:

  1. AE-style GANs: https://github.com/cyclomon/3dbraingen
  2. HA-GAN: https://github.com/batmanlab/HA-GAN
  3. Conditional DPM: https://arxiv.org/abs/2212.08034
  4. MedSyn: https://ieeexplore.ieee.org/document/10566053
  5. MONAI: https://github.com/Project-MONAI/GenerativeModels

Traditional Metrics Evaluation

  1. Generate your own samples
  2. Run evaluation.py or simply use relevant evaluation functions

Anatomical-based Evaluation

To run the 2-stage Anatomical-based evaluation, you need to follow these steps:

  1. Generate certain number of Brain MRI images, and real MRI images. Use Synthseg+ to do a whole brain segmentation for both data, SynthSeg+: https://github.com/BBillot/Synthseg
  2. Run quality control evaluation: Calculate the proportion of each ROI where the quality control exceeds 0.65, and set a threshold to detect the generated results, which we set at 0.95.
  3. Run subcortical evaluation and cortical evalution: Replace the CSV path with the segmentation output of your generated data and the generated data of the real image, and perform regression and Cohen's d calculation. The scripts are run_eval_aseg.py and run_eval_aparc.py, respectively, for subcortical and cortical segmentation results.