EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Adaptive Gated Multimodal Fusion for Explainable and Uncertainty-Aware Tumor Grade Assessment

B. Uthra, R. Sivashankari

IEEE Access · 2026 · 14 · 87392-87413

DOI: 10.1109/access.2026.3701753 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von B. Uthra und 1 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Vellore Institute of Technology, Vellore, India, through the Article Processing Charge.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Deep learning models hold potential progress for automating glioma grading, but their limited interpretability and unreliable confidence estimation hinder clinical adoption. In this work, a trustworthy multimodal deep learning framework is proposed that integrates clinical demographics and genomic mutation profiles to predict glioma grade. A Dynamic Gated Fusion Network is introduced to learn patient-specific weighting between clinical and molecular modalities. To reduce overfitting and derive a clinically meaningful gene panel, Mutual Information-based feature selection is employed, retaining the ten most informative mutations. A systematic pipeline consisting of Monte Carlo Dropout for uncertainty estimation, Temperature Scaling for probability calibration, and SHapley Additive exPlanations for post-hoc explainability was performed to enhance the trustworthiness. The model was evaluated on a publicly available glioma grading dataset obtained from the University of California Irvine repository, derived from The Cancer Genome Atlas, using five-fold stratified cross-validation. The proposed model achieved an accuracy of 0.864, an F1 score of 0.843, and an area under the receiver operating characteristic curve of 0.919, while producing calibrated confidence estimates with an expected calibration error of 0.184. These results highlight the capability of patient-adaptive fusion and reliability-enhancing strategies to bridge the gap between algorithmic performance and real-world clinical deployment.

Förderung

  • Vellore Institute of Technology, Vellore, India, through the Article Processing Charge

Zitieren

APA

Uthra, B., & Sivashankari, R. (2026). Adaptive Gated Multimodal Fusion for Explainable and Uncertainty-Aware Tumor Grade Assessment. IEEE Access, 14, 87392-87413. https://doi.org/10.1109/access.2026.3701753

Vancouver

Uthra B, Sivashankari R. Adaptive Gated Multimodal Fusion for Explainable and Uncertainty-Aware Tumor Grade Assessment. IEEE Access. 2026;14:87392-87413. doi:10.1109/access.2026.3701753

BibTeX
@article{Uthra2026,
  author = {Uthra, B. and Sivashankari, R.},
  title = {Adaptive Gated Multimodal Fusion for Explainable and Uncertainty-Aware Tumor Grade Assessment},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {87392--87413},
  doi = {10.1109/access.2026.3701753}
}
Kostenlos starten →