3765 - Multimodal Magnetic Resonance-Based Clinicoradiomics for Predicting Suspicious Lymph Node Metastasis in Nasopharyngeal Carcinoma

L. Duan¹, Y. Li², S. Zhang³, W. Yin³, L. Zheng⁴, J. Wang⁵, M. Zhang⁵, J. Feng⁵, C. Liang⁵, X. Lv⁴, C. Luo⁴, K. Yuan⁶, K. Xu⁴, Y. Zhou⁷, J. Lang⁸, and J. Yin⁹; ¹Chengdu Medical College, Chengdu, China, ²Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China, Chengdu, China, ³Chengdu Medical College, ChengDu, China, ⁴Department of Radiation Oncology, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, University of Electronic Science and Technology of China, chengdu, China, ⁵Sichuan Cancer Hospital, Chengdu, China, ⁶Sichuan Cancer Hospital & Institute, University of Electronic Science and Technology of China, Chengdu, Sichuan, China, ⁷School of Clinical Medicine & First Affiliated Hospital of Chengdu Medical College, Chengdu, China, ⁸Department of Radiation Oncology, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center; Cancer Hospital affiliate to University of Electronic Science and Technology of China, Chengdu, China, ⁹Sichuan Cancer Hospital & Institute, University of Electronic Science and Technology of China, Chengdu, China

Purpose/Objective(s): Due to the diversity of symptoms and insidious location of Nasopharyngeal carcinoma (NPC),up to 85% of patients have regional lymph node metastasis at the time of diagnosis.The presence of cervical lymph node metastases is an important prognostic factor for patients with NPC; however,the management of patients with cN0 remains controversial.Currently,the clinical consensus is that cervical lymph nodes with a short diameter of >1 cm,necrosis,incomplete peritoneum,and extra-nodal infiltration are metastatic lymph nodes.Suspicious lymph nodes without obvious signs of the above-mentioned characteristics still need to be confirmed using histological or cytological puncture.Consequently we aim to develop a clinical radiomics model using multimodal magnetic resonance to predict the presence of suspicious lymph node metastasis in NPC and validate its accuracy. Materials/

Methods: Retrospectively, a total of 382 lymph nodes from 275 patients with NPC were included in the study(260 in the training set and 122 in the validation set).All lymph nodes, underwent puncture biopsy for pathological evaluation and had complete clinical imaging data.The lymph nodes were grouped based on the patients, and feature selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO) method. The optimal model performance was then constructed using Logistic, Random Forest (RF), XGBoost,and Native Bayes (NB), respectively. Clinical features were further screened using multivariate logistic analysis to build the clinical radiomics model.The results were visualized through column-line diagrams.The model efficacy was assessed using the Area Under the Curve (AUC) and calibration curves of the subjects work characteristics.
Results: In the training set, the stacked sequence-based logistic, RF, XGBoost,and NB imaging genomics models achieved AUCs of 0.810, 0.776, 0.722, and 0.705, respectively. However,the clinical radiomics model incorporating clinical features (short-axis diameter, N stage, lymph node location) showed better calibration and differentiation ability in both the training and validation sets,with an AUC of 0.867 in the training set and 0.840 in the validation set, and clinical decision curves confirmed the higher clinical applicability of the clinical radiomics model.
Conclusion: Clinical radiomics models based on multiple sequences have the potential to more accurately identify metastatic lymph nodes of NPC,even when they do not meet the diagnostic criteria for image-positive lymph nodes,This can significantly impact clinical practice by guiding treatment decisions and improving the precision and cost-effectiveness of treatment protocols. Keywords:Nasopharyngeal carcinoma, radiomics, imaging histology; MRI