Scalable and Efficient Multi-Class Brain Tumor Classification with a Compact Hybrid Deep Learning Model for Real-Time Applications

Medical diagnostics require brain tumor classification to operate in real-time so the task demands accurate results with efficient processing abilities. A new hybrid deep learning solution merges convolutional neural networks (CNNs) with support vector machines (SVMs) to improve classification resu...

Full description

Saved in:
Bibliographic Details
Main Authors: Sohaib R. Awad, Amar I. Daood, Akram A. Dawood
Format: Article
Language:English
Published: Koya University 2025-05-01
Series:ARO-The Scientific Journal of Koya University
Subjects:
Online Access:https://aro.koyauniversity.org/index.php/aro/article/view/2017
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Medical diagnostics require brain tumor classification to operate in real-time so the task demands accurate results with efficient processing abilities. A new hybrid deep learning solution merges convolutional neural networks (CNNs) with support vector machines (SVMs) to improve classification results as this paper describes. A total of four tumor categories including glioma, meningioma, and pituitary tumors together with no tumor appearance contribute to the magnetic resonance imaging (MRI) dataset are used for analysis. We applied and organized three pre-trained deep learning models: Alex-Net, DarkNet-19, and ResNet-50 for comparison. A newly engineered compact CNN model linked with an SVM classifier brought decreased model dimensions while keeping excellent accuracy rates. A proposed compact CNN model delivers 97.50% accuracy through its smaller 2.38 MB size and additional SVM integration results in 97.45% accuracy using 1.43 MB. A Graphical User Interface (GUI) system comprising automated tumor classification capabilities is created to improve real-time systems that visualize MRI scans and illustrate predicted labels in addition to displaying confidence scores. A GUI enables smooth access to the trained model while being suitable for medical practice mobile healthcare environments and edge computing needs. The proposed system shows that lightweight architectures work excellently in real-time system applications especially when used for edge computing and mobile healthcare frameworks. The proposed solution demonstrates superiority over established models through its ability to scale efficiently.
ISSN:2410-9355
2307-549X