Bladder Cancer Multi-Class Segmentation in MRI With Pyramid-In-Pyramid Network

Recognition and segmentation of bladder walls and tumour in MRI is essential for bladder cancer diagnosis. In this paper, we propose a novel Pyramid in Pyramid (PiP) fully convolutional neural network to address this problem. A pyramid backbone with lateral connections between encoder and decoder is utilized to segment the bladder wall and tumour at multiple scales and in an end-to-end fashion. To boost the model's capability of extracting multiscale contextual information, a pyramidal atrous convolution block is embedded into the pyramid backbone. We present experimental results to show that the new method outperforms other state-of-the-art models and that the results have a good consistency with that of experienced radiologists.