Meningioma Detection from MRI: A Comparative Study of Deep Learning Models

Authors

• A. H. Doha, University of southern Queensland, School of Health science and Medical, Australia, 4350Follow
• Sandi Jennifer, University of southern Queensland, School of Health science and Medical, Australia, 4350
• Fatima Alabdaly, University of southern Queensland, School of Health science and Medical, Australia, 4350

Corresponding Author

A. H. Doha

Authors ORCID

A. H. Doha: https://orcid.org/0009-0001-6756-3962

Abstract

This study compares several deep learning models including DenseNet20, CNN, ResNet50, VGG19, EfficientNet-B3, MobileNetV2, NASNetMobile, Xception to automatically classify brain meningiomas from MRI scans. A public dataset was used to evaluate the proposed model. All models were trained, validated, and tested on this dataset. Because meningiomas are common and manual reads, the goal is to speed and standardize diagnosis. Our main contribution is pairing a high-accuracy classifier with deep learning models. Among all models, DenseNet201 performed best, reaching 0.991% accuracy and reliably separating tumorous from non-tumorous MRIs. Future work will expand the dataset and incorporate stronger segmentation to further boost performance and robustness.

Recommended Citation

Doha, A. H.; Jennifer, Sandi; and Alabdaly, Fatima (2026), Meningioma Detection from MRI: A Comparative Study of Deep Learning Models, AUIQ Complementary Biological System: Vol. 3: Iss. 2, 94-100.
DOI: https://doi.org/10.70176/3007-973X.1046
Available at: https://acbs.alayen.edu.iq/journal/vol3/iss2/10

Digital Object Identifier (DOI)

10.70176/3007-973X.1046