Deep Learning-Based Classification of Brain Tumors from

Magnetic Resonance Imaging Scans Using a Convolutional

Neural Network Model

Karim Eldreny1,*

1 Department of Communications and Electronics, Delta Higher Institute of Engineering and Technology, Mansoura 35111, Egypt

Received: December 31, 2025 Revised: February 04, 2026 Accepted: April 01, 2026 ⋆ Corresponding author

ABSTRACT

Brain tumors are serious neurological conditions that require accurate and timely classification to support medical

evaluation and treatment planning. This project presents a deep learning-based system for classifying brain Magnetic

Resonance Imaging (MRI) scans into four categories: glioma, meningioma, pituitary tumor, and no tumor. The

proposed system uses a Convolutional Neural Network (CNN) trained on a balanced dataset of 7,200 MRI images

collected from publicly available sources. The images were preprocessed through RGB conversion, resizing, tensor

transformation, and normalization to ensure consistent input for model training and testing. The trained model

achieved an overall classification accuracy of 94.31% on a held-out test set of 1,600 MRI images, demonstrating

strong performance in multi-class brain tumor classification. A Streamlit-based web application was also developed

to allow users to upload MRI images and view the predicted class, confidence score, and probability distribution

across the four categories. The system is intended for educational and research purposes only and should not replace

professional medical diagnosis, clinical judgment, or radiological evaluation.

Keywords: Brain Tumor Classification; Magnetic Resonance Imaging; Deep Learning; Convolutional Neural

Network; Medical Image Classification

1. INTRODUCTION

Brain tumor diagnosis represents one of the most critical challenges

in contemporary medical imaging because it directly

affects clinical decision-making, treatment planning, surgical

intervention, radiotherapy design, and long-term patient

monitoring. Brain tumors may appear with highly variable

anatomical locations, tissue characteristics, growth patterns,

and visual boundaries, which makes their interpretation from

medical images a complex task even for experienced specialists.

Magnetic Resonance Imaging (MRI) has become a

major imaging modality for brain tumor assessment because

it provides high soft-tissue contrast and can reveal structural

abnormalities within the brain with greater anatomical clarity

than many other imaging techniques. However, the interpretation

of MRI scans remains dependent on expert evaluation,

visual experience, image quality, tumor appearance, and the

availability of trained radiologists. These factors have encouraged

extensive research into artificial intelligence (AI),

machine learning (ML), and deep learning (DL) techniques

that can support automated brain tumor detection and classification

from MRI images [1].

The increasing use of computational intelligence in medical

imaging is mainly driven by the need for accurate, repeatable,

and time-efficient diagnostic support systems. Traditional

clinical interpretation requires careful visual inspection of