A quantitative trigger model to enable personalized adaptive radiotherapy in nasopharyngeal carcinoma
Xiao-li Yu, Jiang Hu, Wen-chao Diao, Guang-yu Wang, Xin Yang, Lu Liu, Xiao-bo Jiang, Chen-di Xu, Liu-wen Lin, Le-cheng Jia, Hua Li, Yan-fei Liu, Yan-ping Mao, Ying Sun, Guan-qun ZhouBackground
Adaptive radiotherapy (ART) is increasingly used for nasopharyngeal carcinoma (NPC), yet optimal triggers remain undefined.
Objectives
This study developed a quantitative trigger model integrating clinical expertise with morphological and dosimetric changes during treatment to support personalized, scalable ART decision-making.
Design
This was a single-center, retrospective cohort study conducted at Sun Yat-sen University Cancer Center.
Methods
Between 2022 and 2024, 60 NPC patients treated on an integrated ART platform (uRT-linac 506c) were enrolled. Forty patients who underwent ART at least once during radiotherapy constituted the training set; morphological and dosimetric parameters between the original treatment plan and fractional radiotherapy were quantified, weighted, and modeled using clinician-simulated ART timing as the reference. The remaining 20 patients (17 with ART and 3 without) served as the validation set.
Results
A total of 1,466 fan-beam CT scans from 60 patients (40 training, 20 validation) were analyzed. A weighted Trigger Score (TS) was developed, combining volumetric and dosimetric scores in target volumes and key organs-at-risk (parotids, submandibular glands, thyroid, brainstem, spinal cord, optic nerves, chiasm, and temporal lobes). Two trigger modes were defined: 1) an absolute TS ≥65 points, or 2) a TS change ratio ≥1.5 (=TS_y × (y - 2)/Σ TS_2: TS_(y-1), where y = fraction number). In the validation cohort, the model showed high concordance with clinician-simulated triggers, achieving ROC-AUCs of 0.879 (daily) and 0.872 (weekly). The model has been integrated into our clinical ART platform for operational use.
Conclusion
We developed and validated a transparent, quantitative trigger model integrating clinical expertise with quantified volumetric/dosimetric changes and structural prioritization to enable individualized ART decision-making for NPC. This model supports automated, real-time ART triggering with adjustable thresholds, adaptable to imaging frequencies, and customizable to institutional practice. However, multicenter or prospective validation is still warranted before widespread clinical adoption.