3662 - Multi-Sequence MRI Based Metachronous Single-Organ Metastasis Prediction of Nasopharyngeal Carcinoma

Y. Huang¹, P. Zhang², M. Feng³, J. Zhou⁴, S. Lu⁵, S. Zhang⁶, W. Wang⁵, Y. Ren⁷, G. Xu⁸, and J. Lang⁶; ¹Department of Radiation Oncology, Radiation Oncology Key Laboratory of Sichuan Province,Sichuan Clinical Research Center for Cancer, Sichuan Cancer Hospital& Institute,Affliated Cancer Hospital of University of Electronic Science and Technology of China, Chengdu, China, ²Department of Radiation Oncology, Sichuan Cancer Hospital& Institution, Sichuan Cancer Center, School of Medicine, University of Electronic Science and Technology of China, Radiation oncology Key Laboratory of Sichuan Province, Chengdu, China, ³The Third Peoples Hospital of Sichuan Province, Chengdu, China, ⁴Sichuan Cancer Hospital & Institute, Sichuan Cancer Center; Cancer Hospital affiliate to University of Electronic Science and Technology of China, Chengdu, China, ⁵Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, School of Medicine, University of Electronic Science and Technology of China, Chengdu, China, ⁶Department of Radiation Oncology, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center; Cancer Hospital affiliate to University of Electronic Science and Technology of China, Chengdu, China, ⁷Sichuan Cancer Hospital and Institute, chengdu, sichuan, China, ⁸Department of Interventional Radiology, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, Radiation Oncology Key Laboratory Of Sichuan Province, Chengdu, China

Purpose/Objective(s): To explore a multi-sequence MRI based prediction of metachronous oligometastases (MO) in nasopharyngeal carcinoma (NPC) patients. Materials/

Methods: The MRI, clinical variable (CV) and follow-up data of the186 patients with NPC were collected. Gross tumor volume (GTV) and lymph node gross tumor volume (GTVln) prior to treatment were defined on T1WI, T2WI and CE-T1WI. After image normalization, the deep learning platform Python (version 3.9.12) was used in Ubuntu 20.04.1 LTS to construct automatic tumor detection and MO prediction model.
Results: We found that the overall performance of automatic tumor detection model was the model based on CE-T1WI. Automatic segmentation algorithm based on Mask Scoring R-CNN had the best overall performance for automatic identification of tumor and metastatic lymph nodes on CE-T1WI images (mAP@0.5=57.8%). When the Mask R-CNN instance segmentation algorithm was used for automatic detection, the AUCs of the MO prediction model based on T1WI, T2WI and CE-T1WI were 0.722, 0.695 and 0.733, respectively. When Cascade Mask R-CNN and Mask Scoring R-CNN instance segmentation algorithm were used for automatic detection, similar prediction model could be acquired. After adding CV, the prediction ability of the prediction model based on T1WI, T2WI and CE-T1WI was further improved under the three automatic segmentation algorithms. The largest AUC (0.775,95% CI 0.606-0.945,p=0.001) was acquired in the prediction model based on the CE-T1WI and CV when Mask R-CNN automatic segmentation algorithm was used. By comparing the 3-year survival of high-risk and low-risk patients based on the combined model, we found that the 3-year DMFS and OS of high and low MO risk patients were 11.4% vs 95% and 85.3% vs 97% respectively (p<0.05). Deep learning based on multimodal MRI is expected to accurately predict MO of NPC.
Conclusion: The intelligent prediction model based on magnetic resonance imaging alone or in combination with clinical data has excellent performance in automatic tumor detection and MO prediction for NPC patients, and is worthy of clinical application.