Capsule GAN for prostate MRI super-resolution

Abstract

Prostate cancer is a prevalent 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. For the first time, MSG-GAN and CapsGAN are utilized simultaneously for high-scale medical 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 substantial margins. For $$8 \times$$ SR, 19.77, 0.60, and 0.79 are achieved for Peak Signal-to-Noise Ratio (PSNR), Structural SIMilarity index metric (SSIM), and Multi-Scale Structural SIMilarity index metric (MS-SSIM), respectively. A new task-specific similarity assessment is introduced as well. A classifier is trained for severe cancer detection. The drop in the accuracy of this model when dealing with super-resolved images is used to evaluate the ability of medical detail reconstruction of the SR models. The proposed model surpassed state-of-the-art work with a 6% margin. The model is also more compact in comparison with the related architecture and has 45% less number of trainable parameters. The proposed SR model is a step towards an efficient and accurate general medical SR platform.