Lung Extraction from CXR

This repository contains 'Lung Extaction System'. It need for efficient computing in registration between previous CXR and post CXR or etc. It is great for extracting 'Lung area' while minimizing loss of information in the original and cutting off unnecessary information. Its performance measured as following.
Precision: 0.788
Recall: 1.0
F1-score: 0.881

Flow Chart

Fig 1. The flowchart of the system.

CNN Structure

Fig 2. The CNN model for classify each regions. It constructed ResNet like model.

Sample of Dataset

Fig 3. Sample data of training dataset. These are reflects shape and texture of data that pre-processed by contouring and masking.

Dataset: Chest X-ray8 from NIH. (Total 112120 CXR images.)

Result

Fig 4. Result of the system. The blue box is ROI of CXR that interested on lung. The red vox is ROI of diagnosis that annotated by NIH.