Machine learning-based prediction of tumor volume changes in MRIdian-based adaptive radiotherapy for prostate cancer
Advances in Radiotherapy & Nuclear Medicine · 15.05.2026 · 4 (2) · 026040003
DOI: 10.36922/arnm026040003 ↗Kurz gefasst
Fachartikel in der Zeitschrift Advances in Radiotherapy & Nuclear Medicine, online erschienen am 15.05.2026. Verfasst von Merve Konuk und 5 weiteren Autor:innen.
Abstract (Originalsprache)
Accurate prediction of inter-fractional tumor volume changes may improve the efficiency and personalization of magnetic resonance-guided adaptive radiotherapy in prostate cancer (PCa). This study investigates the feasibility of using machine learning algorithms to predict inter-fraction gross tumor volume (GTV) changes in PCa during adaptive radiotherapy (ART), aiming to identify the optimal similarity coefficient and model to improve ART decision-making. Retrospective magnetic resonance images from 22 PCa patients treated using the ViewRay MRIdian LINAC system were analyzed. The GTV of each patient was recorded before treatment (GTV0) and during five fractions (GTV1–5). Four different similarity coefficients—dice, Jaccard, Tanimoto, and Ochiai similarity coefficients (DSC, JSC, TSC, and OSC)—were calculated to evaluate inter-fractional GTV variations. Machine learning models—artificial neural networks (ANNs), extreme gradient boosting machine, random forests, classification and regression trees, and k-nearest neighbors—were trained with appropriate hyperparameters to predict tumor volume changes between fractions. Their predictive performances were compared to determine the most effective algorithm. The ANN model demonstrated superior performance in predicting GTV variations across treatment fractions. Among the similarity coefficients, DSC contributed most to predicting inter-fractional tumor volume changes, while OSC had the least impact. The ANN model achieved the highest predictive performance on the independent test dataset (R2 = 0.822). These findings indicate that ANN models can successfully predict inter-fractional GTV variations in PCa, thereby providing valuable support for ART decision-making. Incorporating DSC-based similarity analysis may facilitate ART planning and improve clinical outcomes. The study supports integrating AI-based methods into the ART workflow.
Zitieren
APA
Konuk, M., Kemal, Z., Toker, O., Aras, S., Atalar, B., & İçelli, O. (2026). Machine learning-based prediction of tumor volume changes in MRIdian-based adaptive radiotherapy for prostate cancer. Advances in Radiotherapy & Nuclear Medicine, 4(2), 026040003. https://doi.org/10.36922/arnm026040003
Vancouver
Konuk M, Kemal Z, Toker O, Aras S, Atalar B, İçelli O. Machine learning-based prediction of tumor volume changes in MRIdian-based adaptive radiotherapy for prostate cancer. Advances in Radiotherapy & Nuclear Medicine. 2026;4(2):026040003. doi:10.36922/arnm026040003
BibTeX
@article{Konuk2026,
author = {Konuk, Merve and Kemal, Ziya and Toker, Ozan and Aras, Serhat and Atalar, Banu and İçelli, Orhan},
title = {Machine learning-based prediction of tumor volume changes in MRIdian-based adaptive radiotherapy for prostate cancer},
journal = {Advances in Radiotherapy & Nuclear Medicine},
year = {2026},
volume = {4},
number = {2},
pages = {026040003},
doi = {10.36922/arnm026040003}
}
EUVIMED