Conditional GANs based system for fibrosis detection and quantification in Hematoxylin and Eosin whole slide images. Medical Image Analysis
Manuscript: https://www.sciencedirect.com/science/article/pii/S1361841522001840
Interactive Demo: Coming Back Soon
Cite: Naglah, A., Khalifa, F., El-Baz, A. and Gondim, D., 2022. Conditional GANs based system for fibrosis detection and quantification in Hematoxylin and Eosin whole slide images. Medical Image Analysis, 81, p.102537.
Assessing the degree of liver fibrosis is fundamental for the management of patients with chronic liver disease, in liver transplants procedures, and in general liver disease research. The fibrosis stage is best assessed by histo pathologic evaluation, and Masson’s Trichrome stain (MT) is the stain of choice for this task in many laboratories around the world. However, the most used stain in histopathology is Hematoxylin Eosin (HE) which is cheaper, has a faster turn-around time and is the primary stain routinely used for evaluation of liver specimens. In this paper, we propose a novel digital pathology system that accurately detects and quantifies the footprint of fibrous tissue in HE whole slide images (WSI). The proposed system produces virtual MT images from HE using a deep learning model that learns deep texture patterns associated with collagen fibers. The training pipeline is based on conditional generative adversarial networks (cGAN), which can achieve accurate pixel-level transformation. Our comprehensive training pipeline features an automatic WSI registration algorithm, which qualifies the HE/MT training slides for the cGAN model. Using liver specimens collected during liver transplantation procedures, we conducted a range of experiments to evaluate the detected footprint of selected anatomical features. Our eval uation includes both image similarity and semantic segmentation metrics. The proposed system achieved enhanced results in the experiments with significant improvement over the state-of-the-art CycleGAN learning style, and over direct prediction of fibrosis in HE without having the virtual MT step.
Linux
Python 3.7
See requirements.txt for python packages
This repository containes the codebase for model training, based on TensorFlow implementations of Conditional GANs and Cycle GANs using a range of image sizes (inputs/outputs) to be used for an ensemble learning pipeline similar to what is described in the manuscript.
Step 1: Prepare the data folder (./path/to/data/folder/) as per the following structure
For ConditionalGAN (Paired images for training), each image is horizontally concatenated input/output (side by side)
ROOT/
├──DATASET/
├── train
├── patch_001.jpg
├── patch_002.jpg
└── ...
└── val
├── patch_001.jpg
├── patch_002.jpg
└── ...
└── test
├── patch_001.jpg
├── patch_002.jpg
└── ...
└── monitor
├── patch_001.jpg
├── patch_002.jpg
└── ...For CycleGAN (UnPaired images for training), unpaired images are stored in trainA and trainB folder
ROOT/
├──DATASET/
├── trainA
├── patch_001.png
├── patch_002.png
└── ...
├── trainB
├── patch_001.png
├── patch_002.png
└── ...
└── val
├── patch_001.jpg
├── patch_002.jpg
└── ...
└── test
├── patch_001.jpg
├── patch_002.jpg
└── ...
└── monitor
├── patch_001.jpg
├── patch_002.jpg
└── ...Step 2: Run the following command and adjust the hyper-parameters according to your dataset
python3 ./runHE2MT.py \
--model condGAN1024 \
--epochs 6 \
--lamda 130 \
--monitor_freq 1epoch \
--dataroot ./path/to/data/folder/ \
--experiment_id exp1 \
--modelsave_freq 1 \
--checkpoint_freq 1 Please see runSlurm.sh to run using SLURM and ANACONDA Env on Compute Cluster
Please contact me for any additional questions: [email protected]
Naglah, A., Khalifa, F., El-Baz, A. and Gondim, D., 2022. Conditional GANs based system for fibrosis detection and quantification in Hematoxylin and Eosin whole slide images. Medical Image Analysis, 81, p.102537.
@article{naglah2022conditional,
title={Conditional GANs based system for fibrosis detection and quantification in Hematoxylin and Eosin whole slide images},
author={Naglah, Ahmed and Khalifa, Fahmi and El-Baz, Ayman and Gondim, Dibson},
journal={Medical Image Analysis},
volume={81},
pages={102537},
year={2022},
publisher={Elsevier}
}