Bladder Segmentation in MRI for High Dose Rate Brachytherapy Using Deep Network

Magnetic Resonance (MR) a noninvasive imaging is the benchmark for Image-Guided Brachytherapy (IGBT) as a result of the variance in soft tissue appearance between targets and organs-at-risk (OAR). Due to steep dose gradients in High Dose Rate Brachytherapy (HDRBT) for female gynecological malignancies, identifications of critical organs such as the urinary bladder influence the prescribe dose delivered to the disease and is a significant aspect of high dose rate brachytherapy. In this investigation, we proposed a U-Net-based deep network design for rapid and reproducible automatic shaping of the bladder in MR-guided high dose rate brachytherapy. This investigation evaluated MR images of 40 patients with localized cervical cancer with T2 sequence. We employed a deep convolution neural network, that utilizes both long and short skip connections to enhance the process of feature extraction procedure, which in turn improves the precision of the image partition task. The performance of the modified U-Net-based deep segmentation network is validated using manual annotations by expert radiologists for benchmarking. The intersection over union (IOU) is utilized as the performance evaluation metric for the proposed network. The obtained results prove that the proposed network is very much effective in determining the segmented area properly and its performance is superior compared to the traditional U-Net network.