GAN-Augmented Radiomics and Machine Learning for Post-Therapy GBM Progression Assessment
Journal of Korean Neurosurgical Society · 06.07.2026
DOI: 10.3340/jkns.2026.0075 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Korean Neurosurgical Society, online erschienen am 06.07.2026. Verfasst von Dev Deveswar Rana und 5 weiteren Autor:innen. Beteiligte Einrichtung: Department of Computer Science & Engineering, International Institute of Information Technology, Bhubaneswar, India. Die Arbeit wurde bisher 1-mal zitiert (Crossref).
Abstract (Originalsprache)
Objective
Differentiating true progression (TP) from pseudoprogression (PsP) in post-therapy of glioblastoma multiforme (GBM) remains a critical challenge in neuro-oncology, due to their overlapping radiological characteristics. Consequently, an accurate, non-invasive differentiation between PsP and TP is vital for informing treatment decisions and improving patient outcomes. Therefore, we aimed to present a comprehensive AI-driven framework integrating handcrafted radiomics, deep learning features, and generative modelling to enhance diagnostic precision.
Methods
Multiparametric MRI data (T1, T2, FLAIR, T1GD) from 58 GBM patients (41 TP, 17 PsP) were analysed using three feature extraction strategies: handcrafted radiomics, Vision Transformer (ViT)-based deep features, and a hybrid combination of both. To address class imbalance, synthetic samples were generated using a Conditional Tabular Generative Adversarial Network (CT-GAN). Feature dimensionality was reduced through a two-stage pipeline comprising a Variational Autoencoder (VAE) followed by Principal Component Analysis (PCA), preserving 95% variance. Classifiers, including Support Vector Machine (SVM), Random Forest, XGBoost, and Multi-Layer Perceptron (MLP), were trained and validated via stratified 5-fold cross-validation.
Results
The radiomics-based SVM model yielded the best performance with an accuracy of 91.84% ± 3.59 (95% CI - 0.841 to 0.989) and an AUC of 0.9667 ± 0.0378 (95% CI - 0.920 to 0.998), demonstrating promising performance within the studied dataset and highlighting potential for clinical application, subject to further external validation.
Conclusion
Our demonstrated CT-GAN-augmented hybrid radiomics framework offers a robust, non-invasive, and promising solution for reliably distinguishing PsP from TP. It shows promising performance within the studied dataset, with potential for future clinical translation subject to external validation.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Computer Science & Engineering, International Institute of Information Technology, Bhubaneswar, India.
- Department of Computer Science and Engineering, Indian Institute of Information Technology Vadodara, Vadodara, India.
- Department of Biotechnology and Chemical Engineering, School of Engineering, Faculty of Science, Technology and Architecture, Manipal University Jaipur, Dehmi Kalan, India.
- Infinity Insights Biotechnology Co., Ltd., Taipei, Taiwan.
- Faculty of Health Sciences, Shinawatra University, Pathum Thani, Thailand.
- University of Jember, Jember, Indonesia.
Zitieren
APA
Rana, D. D., Saxena, S., Dubey, N. K., Chaturvedi, H., Lin, J. H., & Yang, Y. S. (2026). GAN-Augmented Radiomics and Machine Learning for Post-Therapy GBM Progression Assessment. Journal of Korean Neurosurgical Society. https://doi.org/10.3340/jkns.2026.0075
Vancouver
Rana DD, Saxena S, Dubey NK, Chaturvedi H, Lin JH, Yang YS. GAN-Augmented Radiomics and Machine Learning for Post-Therapy GBM Progression Assessment. Journal of Korean Neurosurgical Society. 2026. doi:10.3340/jkns.2026.0075
BibTeX
@article{Rana2026,
author = {Rana, Dev Deveswar and Saxena, Sanjay and Dubey, Navneet Kumar and Chaturvedi, Hemang and Lin, Jiann-Her and Yang, Yi-Shan},
title = {GAN-Augmented Radiomics and Machine Learning for Post-Therapy GBM Progression Assessment},
journal = {Journal of Korean Neurosurgical Society},
year = {2026},
doi = {10.3340/jkns.2026.0075}
}
EUVIMED