Semi-supervised model based on implicit neural representation and mutual learning (SIMN) for multi-center nasopharyngeal carcinoma segmentation on MRI

• Published in: Computers in biology and medicine 2024-06, Vol.175, p.108368, Article 108368
• Main Authors: Han, Xu, Chen, Zihang, Lin, Guoyu, Lv, Wenbing, Zheng, Chundan, Lu, Wantong, Sun, Ying, Lu, Lijun
• Format: Article
• Language: English
• Subjects: Adaptive algorithms; Adult; Algorithms; Cancer; Deep Learning; Deep mutual learning; Delineation; Domain adaptive; Female; Hospitals; Humans; Image Interpretation, Computer-Assisted - methods; Image processing; Image segmentation; Implicit neural representation; Labels; Lymph nodes; Machine learning; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Male; Medical imaging; Metastases; Middle Aged; Nasopharyngeal carcinoma; Nasopharyngeal Carcinoma - diagnostic imaging; Nasopharyngeal Neoplasms - diagnostic imaging; Neural coding; Neural Networks, Computer; Patients; Radiation therapy; Representations; Segmentation; Semi-supervised learning; Semi-supervised segmentation; Throat cancer; Tomography; Tumors
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2024.108368

The issue of using deep learning to obtain accurate gross tumor volume (GTV) and metastatic lymph nodes (MLN) segmentation for nasopharyngeal carcinoma (NPC) on heterogeneous magnetic resonance imaging (MRI) images with limited labeling remains unsolved. We collected 918 patients with MRI images from three hospitals to develop and validate models and proposed a semi-supervised framework for the fine delineation of multi-center NPC boundaries by integrating uncertainty-based implicit neural representations named SIMN. The framework utilizes the deep mutual learning approach with CNN and Transformer, incorporating dynamic thresholds. Additionally, domain adaptive algorithms are employed to enhance the performance. SIMN predictions have a high overlap ratio with the ground truth. Under the 20 % labeled cases, for the internal test cohorts, the average DSC in GTV and MLN are 0.7981 and 0.7804, respectively; for external test cohort Wu Zhou Red Cross Hospital, the average DSC in GTV and MLN are 0.7217 and 0.7581, respectively; for external test cohorts First People Hospital of Foshan, the average DSC in GTV and MLN are 0.7004 and 0.7692, respectively. No significant differences are found in DSC, HD95, ASD, and Recall for patients with different clinical categories. Moreover, SIMN outperformed existing classical semi-supervised methods. SIMN showed a highly accurate GTV and MLN segmentation for NPC on multi-center MRI images under Semi-Supervised Learning (SSL), which can easily transfer to other centers without fine-tuning. It suggests that it has the potential to act as a generalized delineation solution for heterogeneous MRI images with limited labels in clinical deployment. •The first MR-based semi-supervised segmentation network for NPC that can be used to segment both GTV and MLN was developed•The CNN was enhanced by equipping with the INR module, which provides a more versatile and inclusive characterization capability for the segmentation network•We synergistically combined Transformer with CNN through mutual learning and added the truncated Gaussian function to the loss function to reduce the consistency weight among CNN output results with low confidence