A Transfer Deep Generative Adversarial Network Model to Synthetic Brain CT Generation from MR Images

Background. The generation of medical images is to convert the existing into one or more required reduce time for sample diagnosis and radiation human body from multiple taken. Therefore, research on has important clinical significance. At present, there are many methods in this field. For example, image process based fuzzy C-means (FCM) clustering method, due unique idea FCM, generated by method uncertain attribution certain organizations. This will cause details be unclear, resulting quality not high. With development generative adversarial network (GAN) model, improved deep GAN model were born. Pix2Pix a UNet. core use paired two types neural fitting, thereby generating high-quality images. disadvantage that requirements data very strict, must one. DualGAN transfer learning. cuts 3D 2D slices, simulates each slice, merges results. every an generated, bar-shaped “shadows” three-dimensional image. Method/Material. To solve above problems ensure generation, paper proposes Dual3D&PatchGAN Since set learning, no need one-to-one sets, only sets needed, which practical significance applications. can eliminate produced DualGAN’s also perform two-way conversion Results. From evaluation indicators experimental results, it analyzed suitable than other models, its effect better.