BCI: Breast Cancer Immunohistochemical Image Generation

Enhanced Research Repository

Original Paper: BCI: Breast Cancer Immunohistochemical Image Generation through Pyramid Pix2pix
Enhanced Implementation: This repository contains a scientifically rigorous implementation of the BCI framework, augmented with Uncertainty Quantification and Perceptual Metrics.

---

Scientific Enhancements

1. Uncertainty Quantification (Clinical Trust)

In medical imaging, a generative model must be trustworthy. We implement Monte Carlo Dropout inference to generate Uncertainty Maps.

• Dark Areas: High confidence (Model is sure).
• Bright Areas: High specific uncertainty (Model is hallucinating or unsure about tissue details). This allows pathologists to identify regions where the synthetic IHC image might be unreliable.

2. Advanced Perceptual Metrics

Standard PSNR/SSIM metrics favor blurry images. We integrate LPIPS (Learned Perceptual Image Patch Similarity) to measure texture fidelity, which is critical for identifying cancerous cells.

3. Reproducible Engineering

• Config-based Experiments: configs/config.yaml replaces messy command-line arguments.
• Package Structure: Modular src/ layout for better maintainability.

---

Quick Start

Installation

git clone https://github.com/mara-werils/BCI-Project.git
cd BCI-Enhanced
pip install -r requirements.txt

Training

Train the model using the configuration file:

python train_net.py --config configs/config.yaml

Evaluation (New Metrics)

Calculate PSNR, SSIM, and LPIPS:

python src/core/evaluate.py --result_path ./results/pyramidpix2pix --gpu

Uncertainty Visualization

Generate Uncertainty Maps for the test set:

python src/core/visualize_uncertainty.py --dataroot ./datasets/BCI --name bci_experiment_v1

Results will be saved in results/bci_experiment_v1/uncertainty/.

---

Repository Structure

├── configs/             # Configuration files
├── src/
│   ├── core/            # Core scripts (train, evaluate, uncertainty)
│   ├── data/            # Dataset loading logic
│   ├── metrics/         # Scientific metrics (LPIPS)
│   ├── models/          # PyTorch models (Pyramid Pix2pix)
│   ├── options/         # Argument parsers
│   └── utils/           # Helper utilities
├── tests/               # Automated tests
└── train_net.py         # Main entry point

Citation

If you use this enhanced repository, please cite the original paper and this implementation:

@inproceedings{Liu_2022_CVPR,
    author    = {Liu, Shengjie and Zhu, Chuang and Xu, Feng and Jia, Xinyu and Shi, Zhongyue and Jin, Mulan},
    title     = {BCI: Breast Cancer Immunohistochemical Image Generation Through Pyramid Pix2pix},
    booktitle = {CVPR Workshops},
    year      = {2022}
}