EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A robust hybrid deep ensemble framework for MRI-based brain tumor classification using power-law preprocessing

Raad Shahamat Alif, Sadia Anjum Mim, Md. Khorshed Alom, Md. Foysal

Computer Methods and Programs in Biomedicine Update · 2026-12 · 10 · 100278

DOI: 10.1016/j.cmpbup.2026.100278 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer Methods and Programs in Biomedicine Update, 2026-12. Verfasst von Raad Shahamat Alif und 3 weiteren Autor:innen. Die Arbeit nennt 50 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate classification of brain tumors using magnetic resonance imaging (MRI) is essential for timely diagnosis and treatment planning. This paper proposes a robust ensemble framework combining deep learning and vision transformer–based models for brain tumor classification, incorporating an efficient Power-Law transformation–based image preprocessing technique to further enhance prediction reliability. The Power-Law transformation, also known as gamma correction, enhances image contrast and structural visibility, thereby facilitating more effective feature extraction by deep models. For performance evaluation, a comparative analysis is performed against traditional preprocessing methods, including Gaussian blur, high-pass filtering, and morphological operations such as erosion and dilation. Specifically, a vision transformer architecture, CoAtNet and four CNN architectures: EfficientNetV2M, DenseNet201, ResNet50, and VGG16, were fine-tuned on the augmented and preprocessed dataset. Two ensemble learning approaches, hard voting and weighted soft voting, were employed to aggregate model predictions and to examine the benefits of combining model predictions. Experimental results show that both ensemble methods achieved slightly higher accuracy than the individual models. Furthermore, it is evident that the weighted soft voting approach with Power-Law–preprocessed images achieved an accuracy of 99.16 %, whereas the hard voting approach with morphologically preprocessed images achieved an accuracy of 97.71 %. Multi-seed and paired statistical analyses showed that weighted soft voting gave the highest mean accuracy, but its advantage over DenseNet201 and over hard voting was small and not statistically significant. Overall, the findings demonstrate the strong performance of the proposed ensemble deep learning framework with Power-Law preprocessing on the evaluated dataset, and indicate its potential as a supporting tool for brain tumor MRI classification, pending further external and clinical validation.

Zitieren

APA

Alif, R. S., Mim, S. A., Alom, M. K., & Foysal, M. (2026). A robust hybrid deep ensemble framework for MRI-based brain tumor classification using power-law preprocessing. Computer Methods and Programs in Biomedicine Update, 10, 100278. https://doi.org/10.1016/j.cmpbup.2026.100278

Vancouver

Alif RS, Mim SA, Alom MK, Foysal M. A robust hybrid deep ensemble framework for MRI-based brain tumor classification using power-law preprocessing. Computer Methods and Programs in Biomedicine Update. 2026;10:100278. doi:10.1016/j.cmpbup.2026.100278

BibTeX
@article{Alif2026,
  author = {Alif, Raad Shahamat and Mim, Sadia Anjum and Alom, Md. Khorshed and Foysal, Md.},
  title = {A robust hybrid deep ensemble framework for MRI-based brain tumor classification using power-law preprocessing},
  journal = {Computer Methods and Programs in Biomedicine Update},
  year = {2026},
  volume = {10},
  pages = {100278},
  doi = {10.1016/j.cmpbup.2026.100278}
}
Kostenlos starten →