ChestX Ray 14 Save Abandoned

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

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.