2ST-UNet: 2-Stage Training Model using U-Net for Pneumothorax Segmentation in Chest X-Rays
Ayat Abedalla, Malak Abdullah, Mahmoud Al‐Ayyoub, Elhadj Benkhelifa
Abstract
Pneumothorax, also called a collapsed lung, is the presence of the air outside of the lung in the space between the lung and chest wall. It is generally diagnosed using a chest X-ray. However, for some cases, the diagnosis can be difficult as other medical conditions appear similarly. Machine Learning algorithms have been providing great assistance in detecting and locating pneumothorax lately. In this paper, we propose a 2-Stage Training system to segment images with pneumothorax. This system has been built based on U-Net, the state-of-the-art Fully Convolutional Network (FCN) architecture, with a backbone Residual Networks (ResNet-34) that is pre-trained on the ImageNet dataset. In the beginning, we train the network at a lower resolution. Then, we load the trained model weights to retrain the network with a higher resolution. Moreover, we utilize different techniques including Stochastic Weight Averaging (SWA), data augmentation, and Test-Time Augmentation (TTA). We use the chest X-ray dataset that is provided by the 2019 SIIM-ACR Pneumothorax Segmentation Challenge, which contains 12047 training images and 3205 testing images. Our experiments show that 2-Stage Training leads to better and faster network convergence. Our method achieves 0.8356 mean Dice coefficient placing it among the top 9% of competitors with a rank of 124 out of 1475.