Generative Adversarial Network Combined with Multitasking to X-ray Cardiac Segmentation

Medical image organ segmentation methods have made unprecedented breakthroughs in recent years. Computer-aided segmentation of medical images can greatly reduce the workload of doctors. Based on the characteristics of chest X-ray images, we proposed improvements over previous Generative Adversarial Network image segmentation methods by jointly a multi-task deep segmentation network for X-ray cardiac segmentation. Dilated convolution has also been added to the generator network to improve segmentation precision. The improved method verifies its performance on the public data set Japanese Society of Radiological Technology (JSRT) and compares it with other well-known medical image segmentation methods. The experimental results verify the effectiveness of the method.