EUVIMEDEUVIMED entdecken
Crossref

Deep Learning–Based Multi-Class Brain Tumor Classification Using an Enhanced CNN Architecture

Mohammed Razia Alangir Banu, Sumit Hazra, A S Gousia Banu

International Journal of Drug Delivery Technology · 14.04.2026 · 16 (9s)

DOI: 10.25258/ijddt.16.9s.77 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 14.04.2026. Verfasst von Mohammed Razia Alangir Banu und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Brain tumours (BTs) are increasing rapidly worldwide and remain a leading cause of death for thousands of individuals each year. Recent advances in computer vision and image processing have significantly improved diagnostic accuracy while reducing costs and processing time, ultimately transforming healthcare technology. Radiologists commonly turn to magnetic resonance imaging (MRI) due to its exceptional sensitivity in detecting even the smallest anomalies within the brain. Over the years, researchers have introduced numerous methods for identifying and classifying brain tumours. Most of these approaches fall into two categories: traditional machine learning (ML) or deep learning (DL). Classical ML classifiers, however, come with a notable drawback—they require manually crafted features, making the process labour-intensive and time-consuming. Deep learning has emerged as a powerful alternative, gaining widespread adoption for both detection and classification tasks. Its primary strength lies in automatic feature extraction, eliminating the need for manual intervention. Building on this foundation, we present DeepTumorNet, a hybrid deep learning model that classifies three distinct types of brain tumours—gliomas, meningiomas, and pituitary tumours—using a convolutional neural network (CNN) architecture. Our journey began with GoogLeNet, a well-established CNN framework. To tailor it specifically for brain tumour classification, we converted it into a hybrid model by removing the final five layers and adding fifteen additional layers. This modification deepened the network and enhanced its learning capacity. Furthermore, we incorporated a leaky ReLU activation function within the feature maps, which improved the model's expressiveness and allowed it to capture more nuanced patterns. To evaluate its effectiveness, we conducted extensive experiments using a contrast-enhanced MRI (CE-MRI) dataset. The results were compelling: DeepTumorNet achieved an accuracy of 96.45%, precision of 95.32%, recall of 96.01%, and an F1 score of 95.20%. These figures consistently outperformed other contemporary approaches applied to the same dataset, underscoring the potential of our hybrid model as a reliable tool for assisting clinicians in accurate brain tumour diagnosis [31].

Zitieren

APA

Banu, M. R. A., Hazra, S., & Banu, A. S. G. (2026). Deep Learning–Based Multi-Class Brain Tumor Classification Using an Enhanced CNN Architecture. International Journal of Drug Delivery Technology, 16(9s). https://doi.org/10.25258/ijddt.16.9s.77

Vancouver

Banu MRA, Hazra S, Banu ASG. Deep Learning–Based Multi-Class Brain Tumor Classification Using an Enhanced CNN Architecture. International Journal of Drug Delivery Technology. 2026;16(9s). doi:10.25258/ijddt.16.9s.77

BibTeX
@article{Banu2026,
  author = {Banu, Mohammed Razia Alangir and Hazra, Sumit and Banu, A S Gousia},
  title = {Deep Learning–Based Multi-Class Brain Tumor Classification Using an Enhanced CNN Architecture},
  journal = {International Journal of Drug Delivery Technology},
  year = {2026},
  volume = {16},
  number = {9s},
  doi = {10.25258/ijddt.16.9s.77}
}
Kostenlos starten →