Chest X-Ray Segmentation

This repository contains PyTorch code train segmentation models on chest-xray images.

View different models, results, and sample images here.

Directory Structure

.
├── data
├── train_seg.py
└── utils
    ├── load_data.py
    ├── load_extra.py
    ├── load_loss.py
    ├── load_model.py
    └── load_train_loop.py

Specifications

Datasets

Dataloaders for following datasets are available. Images should be saved in ./data.

1. Montgomery
2. Shenzhen
3. JSRT

Dataset Modes

Segmentation can be run on the following configurations.

1. Full: Runs on complete dataset.
2. Normal: Runs only on normal chest x-rays.
3. Abnormal: Runs only on abnormal chest x-rays.

Segmentation Models

The following segmentation models are available. Please note that these are custom implementations and don't exeactly match the actual papers.

1. FCN8
2. DeconvNet

Run

Command Blueprint

python train_seg.py --dataset [str] --subset [str]  --model_name [str] --batch_size [int] --epochs [int] --learning_rate [float]

Example Command

python train_seg.py --dataset "montgomery" --subset "full"  --model_name "custom_model_name" --batch_size 16 --epochs 100 --learning_rate 0.01