Conventionally, pneumothorax is diagnosed by a human-driven process in which radiologists and doctors manually review x-ray photographs one by one. This process cannot scale when the number of x-ray photographs is high due to the increase in the number of patients requiring medical treatment. With a limited number of radiologists and doctors, medical centers would not be able to provide diagnosis results for every single case on time. Rushing for diagnosis results and overworking of radiologists and doctors could also affect the quality of diagnosis. With the re-emergence of deep learning that can deal with complex problems at human performance, FPT Software develops an Artificial Intelligent (AI)based technique that enables quick and accurate diagnosis of pneumothorax. We adopt state-of-the-art models such as U-Net, Residual Net, and Capsule Net to develop a new model, namely Residual Capsule Network (RCN). Unlike convolutional neural networks (CNNs) that provide poor translation invariance and lack of information about poses, thus requiring a lot of data for training, RCN is able to preserve information about the pose while achieving a high accuracy (Dice Coefficient) of 83.89% (top 5% of the world) with less training data.