Advanced 3D Dilated Multi-Fiber Network for Real-time Brain tumor segmentation

Abstract

Brain tumor segmentation plays a pivotal role in medical image processing. In this work, we aim to segment brain MRI volumes. 3D convolution neural networks (CNN) such as 3D U-Net [1] and V-net [2] employing 3D convolutions to capture the correlation between adjacent slices have achieved impressive segmentation results. However, these 3D CNN architectures come with high computational overheads due to multiple layers of 3D convolutions, which may make these models prohibitive for practical large-scale applications. 3D DMFNet proposed by Chen Chen [3] reduces computation cost while maintaining high accuracy for brain tumor segmentation. For more improved accuracy keeping efficiency, I propose advanced DMFNet by adjusting new learning rate schedule in training procedure that is called ‘learning rate warmup’ and ‘Cosine Learning Rate Decay’ proposed by Tong He [4]. They examine a collection of training procedure empirically. Therefore, we examined various learning rate schedules to optimize this architecture and achieved dice scores (79.62%, 90.78%, 85.60% for ET, WT and TC, respectively.) which is outperform existing model (3D DMFNet) for WT and TC. And This results also outperform the state-of-the-art algorithm, e.g. NVDLMED [5] for WT.