Synthetic Vertebral Column Fracture Image Generation by Deep Convolution Generative Adversarial Networks

In the field of medical imaging, the challenging objective is to generate synthetic, realistic images which resembles the original images. The generated synthetic images would enhance the accuracy of the computer-assisted classification, Decision Support System, which aid the doctor in diagnosis of diseases. The Generative Adversarial Networks (GANs), is a method of data augmentation which can be used to generate synthetic realistic looking images, however low quality images are generated. For AI models, it is challenging tasks to do classification using this low quality images. In this work, generation of high quality synthetic medical image using Deep Convolutional Generative Adversarial Networks (DCGANs) is presented. Data augmentation method by DCGANs is illustrated on the limited dataset of CT (Computed Tomography) images of vertebral column fracture. A total of 340 CT scan images were taken for the study, which comprises of complete burst fracture scans of vertebral column. The evaluation of the generated images was done with Visual Turing Test.