normalct/chestx-ray-14

Segmentation Guided Thoracic Classification

Lung Segmentation and Thoracic Classification using Deep Learning

• This is my project for Thoracic Classification using explicit attention mechanism by lung segmentation on ChestX-ray-14 dataset for 14 different lung diseases (i.e. Atelectasis, Cardiomegaly, Effusion, Infiltration, Mass, Nodule, Pneumonia, Pneumothorax, Consolidation, Edema, Emphysema, Fibrosis, Pleural Thickening, and Hernia).
• Segmentation Guided Thoracic Classification.

Architecture

• Lung segmentation use a modified version of ResNet with U-net liked decoder path.
• Thoracic classification use DenseNet-121.

Prerequisites

• Python 3.6+ (I use f-format string)
• Pytorch 1.0.0
• Fastai, I used this branch and create symbolic link from fastai/old/fastai to src/fastai directory
• CheXNet and Unet models are in models directory
• Start by looking at src/chexnet.py, src/unet.py and src/server.py

Dataset

• NIH ChestX-ray Dataset with 112,120 images from 30,805 patients for classification, the dataset can be downloaded here.
• JSRT (Japanese Society of Radiological Technology) and MC (Montgomery County) for lung segmentation. JSRT can be downloaded here with SCR mask, MC dataset can be downloaded here.

Result

• Lung Segmentation result on Top: JSRT dataset, Bottom: MC dataset. The green contour indicates the ground truth and the red mask indicates the predicted mask. Ω denotes the IoU value.

• Lung segmentation on ChestX-ray 14 dataset.

• Comparision

Method	Atel	Card	Effu	Infi	Mass	Nodu	Pne1	Pne2	Cons	Edem	Emp	Fibr	PT	Hern	Mean
Yao et la.	0.772	0.904	0.859	0.695	0.792	0.717	0.713	0.841	0.788	0.882	0.829	0.767	0.765	0.914	0.803
Rajpurkar et la.	0.809	0.925	0.864	0.735	0.867	0.78	0.768	0.889	0.79	0.888	0.937	0.805	0.806	0.916	0.841
Guan et la.	0.853	0.939	0.903	0.754	0.902	0.828	0.774	0.921	0.842	0.924	0.932	0.864	0.837	0.921	0.871
Mine	0.831	0.918	0.883	0.712	0.859	0.789	0.765	0.88	0.813	0.899	0.911	0.826	0.782	0.943	0.843

Each pathology is denoted with its first four characteristics, e.g., Atelectasis with Atel. Pneumonia and Pneumothorax are denoted as Pneu1 and Pneu2, respectively. PT represents Pleural Thickening

Demo

Reference

• X. Wang, Y. Peng, L. Lu, Z. Lu, M. Bagheri, and R. M. Summers, “ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases,” ArXiv170502315 Cs, May 2017.
• P. Rajpurkar et al., “CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning,” ArXiv171105225 Cs Stat, Nov. 2017.
• L. Yao, E. Poblenz, D. Dagunts, B. Covington, D. Bernard, and K. Lyman, “Learning to diagnose from scratch by exploiting dependencies among labels,” ArXiv171010501 Cs, Oct. 2017.
• Q. Guan, Y. Huang, Z. Zhong, Z. Zheng, L. Zheng, and Y. Yang, “Diagnose like a Radiologist: Attention Guided Convolutional Neural Network for Thorax Disease Classification,” Jan. 2018.