DDNet: 3D densely connected convolutional networks with feature pyramids for nasopharyngeal carcinoma segmentation

Abstract

Radiation therapy is the standard treatment for early stage Nasopharyngeal cancer (NPC). Thus, accurate delineation of target volumes at risk in NPC is important. While manual delineation is time-consuming and labour-intensive process and also leads to significant inter-and intra-practitioner variability. Thus, computer-aided segmentation algorithm is required. However, segmentation task is not trivial due to large variations (e.g., shape and size) of nasopharynx structure across subjects. Moreover, extreme foreground and background class imbalance in NPC segmentation remains challenge. In this paper, we propose a threedimensional densely connected convolutional neural network with multi-scale feature pyramids for NPC segmentation. We adapt the densely connected convolutional block into a new structure via adding feature pyramids. The concatenated pyramid feature carries multi-scale and hierarchical semantic information which is effective for segmenting different size of tumors and perceiving hierarchical context information. To address the foreground and background imbalance problem, we propose an enhanced version of focal loss. It prevents the large number of negative voxels far from boundaries from overwhelming the segmentation algorithm. We validated the proposed method on 120 clinical subjects. Experimental results demonstrate that our approach out-performed state-of-the-art methods and human experts.

BSMM-Net: Multi-modal neural network based on bilateral symmetry for nasopharyngeal carcinoma segmentation

“…The difference in performance between SwinUnet and TransUnet proves that the long-range spatial relationship including the bilateral symmetry of the head is a key for detecting NPC lesions. Although resnet34 ( Li et al, 2021 ) used as an encoder by BASNet is pre-trained in the ImageNet datasets, the dataset is not large enough to train a large number of parameters of BASNet. Therefore, the BASNet achieved the worst performance in the experiments.…”

“… Guo et al (2020) utilized low-level features and multi-scale information to improve delineation by applying long-range skip connection and multi-scale feature pyramid. For multi-scale and multi-level information, DDNet introduces dense connections and feature pyramids to the networks and achieves excellent performance in NPC MRI ( Li et al, 2021 ). DA-DSUnet exhibits an improved performance by utilizing multi-level features based on channel attention and position attention mechanisms ( Tang et al, 2021 ).…”

DMCT-Net: dual modules convolution transformer network for head and neck tumor segmentation in PET/CT

“…The two-stage strategy alleviated the problem that the contrast of CT image was low, and it was difficult to locate and distinguish tumors, but it was more complex than the single-stage strategy. Li et al (2022) combined densely connected convolutional blocks and multi-scale feature pyramids to construct new network structures. This structure utilized multi-scale semantic information to effectively segment tumors, but the extraction and utilization of global features were insufficient.…”

CAFS: An Attention-Based Co-Segmentation Semi-Supervised Method for Nasopharyngeal Carcinoma Segmentation

“…The most common method for the automatic segmentation of nasopharyngeal carcinoma is the fully-supervised methods [24][25][26][27][28]. In the last few decades, deep learning methods have been increasingly used in medical image segmentation [29][30][31].…”