Comparative evaluation of lightweight CNN and YOLOv8 models for brain tumor detection in resource-constrained settings

Ismail, Ahsiah and Idna Idris, Mohd Yamani (2026) Comparative evaluation of lightweight CNN and YOLOv8 models for brain tumor detection in resource-constrained settings. International Journal on Perceptive and Cognitive Computing, 12 (1). pp. 53-64. E-ISSN 2462-229X

Preview

PDF - Published Version Download (4MB) | Preview

Abstract

Brain tumor detection is essential for timely diagnosis, early intervention, and effective treatment planning. With advancements in artificial intelligence (AI), deep learning methods have emerged as powerful tools in medical imaging, offering automated, consistent, and efficient detection of brain abnormalities. However, achieving clinically reliable performance requires models that can accurately differentiate between tumor and non-tumor cases. This paper investigates and compares the performance of two deep learning models which are a lightweight Convolutional Neural Network (CNN) and the You Only Look Once (YOLO) YOLOv8 model for brain tumor classification in resource-constrained setting. Both models were trained and evaluated using the BR35H dataset, which comprises 3,000 MRI scans categorized into tumor and non-tumor classes. The performance of the models are evaluated using accuracy, precision, recall, F1-score as well as inference time supplemented by confusion matrix, ROC analysis and Grad-CAM visualizations to assess class-wise prediction performance. The experimental results indicate that YOLOv8 demonstrated high predictions across both tumor and non-tumor categories. YOLOv8 outperformed the CNN, achieving an accuracy of 0.998, precision of 0.997, recall of 1.00, and an F1-score of 0.998. However, only a minimal difference was observed in the inference time per image between YOLOv8 and the CNN, with YOLOv8 being slower by just 10.6 ms. Finally, the results demonstrate YOLOv8’s robustness and reliability for early tumor detection, a critical factor in preventing diagnostic delays. The findings further highlight YOLOv8’s suitability for integration into clinical decision-support systems, particularly in resource-constrained environments where accurate and fast automated diagnosis can significantly enhance patient care

Item Type:	Article (Journal)
Uncontrolled Keywords:	deep learning, CNN, YOLO, Brain tumor detection
Subjects:	T Technology > T Technology (General)
Kulliyyahs/Centres/Divisions/Institutes (Can select more than one option. Press CONTROL button):	Kulliyyah of Information and Communication Technology Kulliyyah of Information and Communication Technology
Depositing User:	Dr AHSIAH ISMAIL
Date Deposited:	19 Feb 2026 11:06
Last Modified:	19 Feb 2026 11:06
Queue Number:	2026-02-Q2194
URI:	http://irep.iium.edu.my/id/eprint/127488

Actions (login required)

View Item