EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Predicting early side effects in prostate stereotactic body radiation therapy using a recurrent neural network incorporating longitudinal anatomy-specific dose metrics

Yunkui Pang, Jonathan Pham, Amar Kishan, Luca F. Valle, Michael Steinberg, Michael Repka, Ari Wijetunga, Shivani Sud, Zachary Gude, Pew-Thian Yap, Jun Lian, X. Sharon Qi

Physics and Imaging in Radiation Oncology · 2026-11 · 101119

DOI: 10.1016/j.phro.2026.101119 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Physics and Imaging in Radiation Oncology, 2026-11. Verfasst von Yunkui Pang und 11 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background and Purpose: Stereotactic body radiotherapy (SBRT) is widely used for localized prostate cancer, yet genitourinary (GU) and gastrointestinal (GI) adverse effects remain important determinants of quality of life. This study evaluated whether incorporating fraction-specific delivered dose metrics improves prediction of early treatment-related side effects. Materials and methods: We developed a novel recurrent neural network (RNN)-based framework that incorporated longitudinal fraction-specific dose metrics to predict acute GU/GI side effects. The framework was evaluated with 69 prostate patients treated with magnetic resonance-guided SBRT. Dose-volume histograms from the initial plan and fraction-specific dose metrics were derived for model prediction. Model performance was evaluated using five-fold nested cross-validation, with 55 patients for training and 14 for validation in each outer fold. Within each training set, three-fold inner cross-validation was used for feature selection and hyperparameter optimization. Predictive performance was measured by area-under-the-curve (AUC), and differences between models were compared using DeLong tests.

Results

Using all treatment fractions performed similarly to the planning dose for predicting acute GU grade ≥ 2 side effects (AUC, 0.58 vs 0.56; p = 0.37). Incorporating fractions 1 and 2 improved the AUC to 0.74 ( p = 0.001) with the planning dose, and 0.76 without it ( p = 0.001). Early fraction-specific dose metrics also improved prediction of GU grade ≥ 1 and GI grade ≥ 1 side effects. The leading features were urethra V 41.5 Gy and trigone V 39.8 Gy for GU side effects.

Conclusions

RNN-based longitudinal dose metrics modeling improved the prediction of early GU/GI side effects and may facilitate personalized, side effect-aware adaptive radiotherapy for prostate SBRT.

Zitieren

APA

Pang, Y., Pham, J., Kishan, A., Valle, L. F., Steinberg, M., Repka, M., Wijetunga, A., Sud, S., Gude, Z., Yap, P., Lian, J., & Qi, X. S. (2026). Predicting early side effects in prostate stereotactic body radiation therapy using a recurrent neural network incorporating longitudinal anatomy-specific dose metrics. Physics and Imaging in Radiation Oncology, 101119. https://doi.org/10.1016/j.phro.2026.101119

Vancouver

Pang Y, Pham J, Kishan A, Valle LF, Steinberg M, Repka M, et al. Predicting early side effects in prostate stereotactic body radiation therapy using a recurrent neural network incorporating longitudinal anatomy-specific dose metrics. Physics and Imaging in Radiation Oncology. 2026:101119. doi:10.1016/j.phro.2026.101119

BibTeX
@article{Pang2026,
  author = {Pang, Yunkui and Pham, Jonathan and Kishan, Amar and Valle, Luca F. and Steinberg, Michael and Repka, Michael and Wijetunga, Ari and Sud, Shivani and Gude, Zachary and Yap, Pew-Thian and Lian, Jun and Qi, X. Sharon},
  title = {Predicting early side effects in prostate stereotactic body radiation therapy using a recurrent neural network incorporating longitudinal anatomy-specific dose metrics},
  journal = {Physics and Imaging in Radiation Oncology},
  year = {2026},
  pages = {101119},
  doi = {10.1016/j.phro.2026.101119}
}
Kostenlos starten →