论文部分内容阅读
Prostate cancer is the most common cancer in males and a major cause of cancer-related death. Magnetic resonance (MR) imaging is recently emerging as a powerful tool for prostate cancer diagnosis. To clinically diagnose prostate cancer, doctors need to segment the prostate area in the MR image. However, manual segmentation is time consuming and influenced by the physician’s experience. Computer-aided diagnosis and decision-making systems have shown great effectiveness in assisting doctors for this purpose. At the same time, deep leaing based on Generative Adversarial Networks can be applied to the segmentation of prostate MR images. In this paper, we propose a new computer-aided diagnosis and decision-making system based on a deep leaing model to automatically segment the prostate region from prostate MR images. Additionally, receptive field block (RFB) was integrated into the model to enhance the discriminability and robustness of the extracted multi-scale features. We also introduced dense upsampling convolution instead of the traditional bilinear interpolation to capture and recover fine-detailed information. Adversarial training was used to train the model, and the segmentation results were experimentally tested. The results showed that adversarial training and RFB are indeed effective, and the proposed method is superior to other methods on various evaluation metrics.