Pneumothorax Segmentation with Effective Conditioned Post-Processing in Chest X-Ray

The automatic detection of abnormal elements in chest Xrays (CXR), such as pneumothorax, is important and challenging problem. Screening for unexpected findings or any surveys in the complicated conditions are the common scenarios for the radiologists in their clinical workflow, where the automated solutions are required. The pneumothorax can be caused by a blunt chest injury, damage from underlying lung disease or it may occur for no obvious reason at all [1]. This is one of the complex problems for the experts manual detection, which can be solved automatically and simplify the clinical workflow. Proposed method presents new segmentation pipeline for the CXR images with the multi step conditi oned post-processing. This approach leads to the significant improvement compare with any ”baseline” by the reduction of the totally missed and false positive detections of the pneumothorax collapse regions. Obtained results demonstrate very high accuracy and strong robustness due to very similar performance on the double-stage test dataset. Final Dice scores are 0.8821 and 0.8614 for ”stage 1” and ”stage 2” test datasets respectively, what is resulted in top 0.01% standing of the private leaderboard on the Kaggle competition platform. Code is available at https://github.com/n01z3/kaggle-pneumothoraxsegmentation.