Main Session
Sep
28
PQA 03 - Digital Health Innovation and Informatics, Patient Safety & Quality, and Radiation and Cancer Biology
2676 - Longitudinal and Multimodal Magnetic Resonance Imaging-Based Radiomics for Predicting Pathologic Complete Response to Neoadjuvant Chemoimmunotherapy in Head and Neck Squamous Cell Carcinoma
Presenter(s)
Ya-Ni Zhang, MS - Sun Yat-Sen University Cancer Center, Guang Dong Province, Guangdong
Y. N. Zhang1, J. X. Lin2, Y. Liu1, H. Li3, X. Liu2, F. Y. Xie1, F. Han1, X. Liu4, Q. Zhou2, X. W. Deng1, Y. Peng1, and C. Y. Chen1; 1Department of Radiation Oncology, Sun Yat-sen University Cancer Center, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, China, Guangzhou, China, 2Manteia Technologies Co Ltd, Xiamen, Fujian, China, 3Department of Radiology, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong, China, Guangzhou, China, 4Department of Head and Neck Surgery, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong, China, Guangzhou, China
Purpose/Objective(s):
Neoadjuvant treatment is essential for reducing tumor burden, enhancing resection rates, and improving outcomes in locally advanced head and neck squamous cell carcinoma (HNSCC). However, the variability in patient responses underscores the need for precise preoperative predictions of treatment responses. We aimed to identify and validate magnetic resonance imaging (MRI)-based markers from pre- and post-neoadjuvant chemoimmunotherapy to develop a radiomic model for predicting pathologic complete response (pCR).Materials/Methods:
Between January 2019 and December 2023, we conducted a study involving 159 HNSCC patients at one institution, all of whom received neoadjuvant chemoimmunotherapy followed by surgical intervention. Our analysis encompassed clinical data, pre- and post-treatment MRI scans, and pathology results. We evaluated single-modal and multimodal models using features derived from pre-treatment, post-treatment, and delta changes. Radiomic model performance was assessed through the area under the receiver operating characteristic curve (AUC), with data analysis planned from February to December 2025.Results:
Among 159 patients (126 [79.2%] men; median [range] age, 55 [24–82] years), pCR rates were 65.6% and 64.5% in training and test cohorts, respectively. The XGBoost model integrating multiparametric MRI with clinical data demonstrated optimal performance among the evaluated models. The radiomic signature showed strong predictive capability for pCR (AUCs: 0.845, 0.851, and 0.959 in pre-treatment, post-treatment, and delta models, respectively, in the test cohort). Decision-curve analysis confirmed clinical utility. The analysis identified four significant longitudinal radiomic features associated with PD-L1 expression levels, effectively discriminating combined positive-score groups. While pre-treatment/delta features showed transient associations with progression-free survival in low-risk training subsets, these were not sustained post-treatment or observed in the test cohort.Conclusion:
Integration of multi-time-point delta radiomic features with multimodal data represents a potential clinical tool for preoperative prediction of therapeutic response in HNSCC, enabling timely, personalized treatment planning for patients unlikely to respond.