Benchmarking Deep Convolutional Neural Networks for Brain Tumor Detection Using Magnetic Resonance Imaging Data
CommIT (Communication and Information Technology) Journal · 08.09.2026 · 415
DOI: 10.21512/commit.v20i2.14392 ↗Kurz gefasst
Fachartikel in der Zeitschrift CommIT (Communication and Information Technology) Journal, online erschienen am 08.09.2026. Verfasst von Tegar Anugrah Firdaus und 8 weiteren Autor:innen. Die Arbeit nennt 66 Literaturangaben.
Abstract (Originalsprache)
Brain tumors are associated with high mortality and severe neurological effects, making accurate and timely diagnosis essential. However, manual interpretation of Magnetic Resonance Imaging (MRI) remains subjective and time-consuming. Deep learning, particularly Convolutional Neural Networks (CNNs), offers a promising approach to improving diagnostic accuracy and efficiency. This research systematically benchmarks five CNN architectures (VGG19, DenseNet201, ResNet50, Inception-v3, and MobileNet) on balanced and naturally imbalanced MRI datasets. Its novelty lies in evaluating all models under a unified preprocessing and transferlearning framework across both dataset distributions, with Grad-CAM used to support interpretability. Preprocessing includes resizing, adaptive masking, Gaussian blur, Contrast Limited Adaptive Histogram Equalization (CLAHE), normalization, and augmentation, followed by transfer learning using ImageNet-pretrained models. Performance is evaluated using accuracy, precision, recall, specificity, F1-score, and Area Under the Curve (AUC). Results show that VGG19 achieves the highest accuracy (0.937 for balanced and 0.962 for imbalanced data), recall (0.991 and 0.984, respectively), and F1-score, demonstrating strong sensitivity and generalization. DenseNet201 achieves the highest AUC (0.992 and 0.996) and high specificity, indicating strong overall discriminative performance. These findings suggest that VGG19 is particularly suitable for sensitivity-oriented early screening, whereas DenseNet201 is advantageous when overall discrimination and specificity are prioritized. In conclusion, CNN-based models show substantial potential for brain tumor classification, with architecture selection depending on specific clinical priorities.
Zitieren
APA
Firdaus, T. A., Rais, B., Salim, M. J., Negara, A. R., Zia, M. Q. Z., Thwe, Y., Uda, M. N. A., Hashim, U., & Pamungkas, Y. (2026). Benchmarking Deep Convolutional Neural Networks for Brain Tumor Detection Using Magnetic Resonance Imaging Data. CommIT (Communication and Information Technology) Journal, 415. https://doi.org/10.21512/commit.v20i2.14392
Vancouver
Firdaus TA, Rais B, Salim MJ, Negara AR, Zia MQZ, Thwe Y, et al. Benchmarking Deep Convolutional Neural Networks for Brain Tumor Detection Using Magnetic Resonance Imaging Data. CommIT (Communication and Information Technology) Journal. 2026:415. doi:10.21512/commit.v20i2.14392
BibTeX
@article{Firdaus2026,
author = {Firdaus, Tegar Anugrah and Rais, Bryan and Salim, Marcelinus Jonathan and Negara, Andhika Rastra and Zia, Muhammad Qaiser Zaydan and Thwe, Yamin and Uda, Muhammad Nur Afnan and Hashim, Uda and Pamungkas, Yuri},
title = {Benchmarking Deep Convolutional Neural Networks for Brain Tumor Detection Using Magnetic Resonance Imaging Data},
journal = {CommIT (Communication and Information Technology) Journal},
year = {2026},
pages = {415},
doi = {10.21512/commit.v20i2.14392}
}
EUVIMED