Capsule GAN for Prostate MRI Super-Resolution

Prostate cancer is a very common disease among adult men. One in seven Canadian men is diagnosed with this cancer in their lifetime. Super-Resolution (SR) can facilitate early diagnosis and potentially save many lives. In this paper, a robust and accurate model is proposed for prostate MRI SR. The model is trained on the Prostate-Diagnosis and PROSTATEx datasets. The proposed model outperformed the state-of-the-art prostate SR model in all similarity metrics with notable margins. A new task-specific similarity assessment is introduced as well. A classifier is trained for severe cancer detection and the drop in the accuracy of this model when dealing with super-resolved images is used for evaluating the ability of medical detail reconstruction of the SR models. The proposed SR model is a step towards an efficient and accurate general medical SR platform.