Advanced machine learning strategies for predicting therapy response in preclinical glioblastoma using longitudinal MRI

Glioblastoma (GB) is the most aggressive primary brain tumor, characterized by a poor prognosis, limited response to therapy, and high rates of recurrence. Early therapeutic response assessment is challenging due to phenomena such as pseudoresponse and pseudoprogression. This study explores the potential of advanced machine learning (ML) strategies to predict long-term therapy outcomes using longitudinal T2weighted Magnetic Resonance Imaging (MRI) data from a preclinical GL261 glioblastoma mouse model, acquired prior and during treatment.

We compare two distinct approaches: a classical pipeline based on radiomic features coupled with an XGBoost classifier, and a deep learning (DL) pipeline using a fine-tuned EfficientNetB0 model. Our results demonstrate that while the radiomics approach identifies interpretable imaging biomarkers and achieves good predictive performance (AUC ≈ 0.770, 95% CI 0.703–0.832), the DL-based model outperforms it across most evaluation metrics, reaching an AUC of 0.868 (95% CI 0.810– 0.918) and a sensitivity of 0.818. The DL model shows better generalization across individual subjects, and the discriminative performance improves progressively throughout the follow-up period for both approaches.

Interpretability analysis via Grad-CAM confirms that the DL model’s predictions are driven by anatomically relevant features. These findings suggest that DL, enhanced by transfer learning, has the potential to serve as a powerful non-invasive tool for the early prediction of treatment efficacy in glioblastoma, paving the way for more robust and personalized therapy monitoring. The authors thank the joint preclinical MRI facility of the Universitat Autònoma de Barcelona and the Centro de Investigación Biomédica en Red-Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), where all MRIexperiments were carried out, as part of Unit 25 of NANBIOSIS (Cerdanyola del Vallès, Spain).

This work was funded by MICIU/AEI/10.13039/501100011033/ (PID2023-147750NB-I00 to A.P.C.) and MICIU/AEI/10.13039/501100011033/FEDER (PID2022-143299OB-I00 to A.V.), Generalitat de Catalunya (2021 SGR 0135), and Instituto de Salud Carlos III (CIBER-BBN group CB06-01-0010). Department of Computer Science, Universitat Politècnica de Catalunya (UPC), Barcelona, Spain Department of Biochemistry and Molecular Biology, Universitat Autònoma de Barcelona (UAB), Cerdanyola del Vallès, Barcelona, Spain Centro de Investigación Biomédica en Red en Bioingeniería, Biomateriales y Nanomedicina (CIBER- BBN), Instituto de Salud Carlos III, Madrid, Spain Intelligent Data Science and Artificial Intelligence (IDEAI) Research Center, UPC, Barcelona, Spain Correspondence to Ana Paula Candiota or Alfredo Vellido. The authors declare no competing interests.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Below is the link to the electronic supplementary material. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made.

The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.