Hybrid Deep Learning Framework with Explainable AI for Multi-Class Brain Tumor Classification Using MRI Images
Anbar Journal of Engineering Sciences · 20.05.2026 · 17 (1) · 192-203
DOI: 10.37649/ajes.26.0032 ↗Kurz gefasst
Fachartikel in der Zeitschrift Anbar Journal of Engineering Sciences, online erschienen am 20.05.2026. Verfasst von Hamid R. Alsanad.
Abstract (Originalsprache)
Brain tumor classification from MRI is essential for efficient diagnosis and treatment. In this study, the authors proposed a hybrid deep learning model based on CNN features and GLCM texture-mapped image descriptors for brain tumor multi-class classification. The proposed framework follows a three-stage pipeline: deep feature extraction using a pre-trained ResNet-18 network, texture feature fusion via Gray-level Co-occurrence Matrix (GLCM) descriptors, and final classification through a fully connected softmax layer. Additional classifiers including an ECOC based SVM,.were evaluated solely as baseline comparators to contextualise the performance of the proposed architecture. The dataset has 7200 MRI images categorized as glioma, meningioma, pituitary tumor, and no tumor. The data is split into balanced training/ testing sets of 6400/ 1600 scans. Experiments on the dataset show that this hybrid model achieves an accuracy of 96.8%, around 4.5% better than conventional CNN-only methods. The model gets precision of 96.5%, recall, and the F score at 96.2% and 96.3% respectively. Class-wise analysis revealed that the method performed relatively well across the pituitary (AUC = 0.994) and no-tumor (AUC = 0.985) categories, while the glioma class was the hardest to predict, with the highest AUC of 0.888, thus the class appeared to be heterogeneous in nature. The GLCM features improved robustness by summarizing subtle changes in texture features that were not fully captured by deep features in isolation. The further addition of Grad-CAM added a layer of visual interpretability by providing heat maps indicating relevant regions of the tumor, thereby improving the reliability of the model in practice.
Zitieren
APA
Alsanad, H. R. (2026). Hybrid Deep Learning Framework with Explainable AI for Multi-Class Brain Tumor Classification Using MRI Images. Anbar Journal of Engineering Sciences, 17(1), 192-203. https://doi.org/10.37649/ajes.26.0032
Vancouver
Alsanad HR. Hybrid Deep Learning Framework with Explainable AI for Multi-Class Brain Tumor Classification Using MRI Images. Anbar Journal of Engineering Sciences. 2026;17(1):192-203. doi:10.37649/ajes.26.0032
BibTeX
@article{Alsanad2026,
author = {Alsanad, Hamid R.},
title = {Hybrid Deep Learning Framework with Explainable AI for Multi-Class Brain Tumor Classification Using MRI Images},
journal = {Anbar Journal of Engineering Sciences},
year = {2026},
volume = {17},
number = {1},
pages = {192--203},
doi = {10.37649/ajes.26.0032}
}
EUVIMED