Enhancing Brain Tumor Detection with Deep Learning Models: A Comparative Analysis

Mathai, Asish Mathai (2024) Enhancing Brain Tumor Detection with Deep Learning Models: A Comparative Analysis. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (892kB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Medical imaging for the detection of brain tumors has lately been significantly enhanced by the inclusion of deep learning technologies. However, detailed comparative research into the clinical practicality of state-of-the-art models has been scant. This is a comparative study of three deeper architectures of convolutional neural networks, namely YOLOv9, PaliGemma, and Detectron2, in detecting brain tumors based on their performance metrics concerning accuracy, speed of processing, computational efficiency, and clinical applicability. Implementing and testing each model with the standardized protocols in this work, the dataset used contained 8,903 brain MRI images covering four categories: with no tumor, meningioma, pituitary, and glioma. The results obtained indicate that YOLOv9 topped with an mAP50 of 0.958 and mAP50-95 of 0.78, leveraging far in front of any results obtained with Detectron2 with a mAP50 of 0.698, and PaliGemma at a mAP50 of 0.482. Although PaliGemma introduces a very unique approach of vision-language, its mediocre performance indicates that domain-specific optimizations are required. Detection performance: Overall, Detectron2 has shown great specialisation capability. It seems particularly good in detecting meningioma (76.86% AP). These findings give evidence-based recommendations for the choice of models toward specific clinical needs and further contribute to the advance of AI-assisted medical imaging. The study illustrates the potential clinical implementation of YOLOv9 while pointing toward a future direction for hybrid architectures that model strengths of traditional object detection with advanced language understanding relative to specific tasks. This comprehensive evaluation lays the ground for further developments in automated brain tumor detection systems.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Tomer, Vikas UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning Q Science > Life sciences > Medical sciences > Pathology > Tumors
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	03 Sep 2025 13:59
Last Modified:	03 Sep 2025 13:59
URI:	https://norma.ncirl.ie/id/eprint/8748

Actions (login required)

View Item