Brain Tumour Prediction Using CNN Algorithm

Athanti Gurunathan, Susanth Kumar (2025) Brain Tumour Prediction Using CNN Algorithm. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (606kB) | Preview

Abstract

Diagnosis of brain tumor is a crucial problem in modern neurology because the timely and correct diagnosis of a brain tumor would make a substantial contribution to the survival of a patient and his/her medical rehabilitation. Manual MRI analysis is slow, has human error and a problem with inter-observer variation, and thus automated diagnostic solutions have an acute demand. The research question of this study is: How do Architecture and preprocessing techniques combination relating to CNNs provide the best solution to the accurate classification of brain tumors based on MRI images?

This evaluated four different CNN models based on Brain Tumor Classification MRI Kaggle database (6,056 scans of MRI with three different tumors: glioma, meningioma and pituitary tumors). The framework of our tests pitted a simple CNN model, a slightly modified CNN version through the use of the batch normalization technique, a ResNet50 transfer learning solution, and a new custom CNN-based solution inspired by the YoloV8 architecture. The preprocessing of each model involved systematic operations such as scaling of the images, normalization of the pixel values and data augmentation methods.

The CNN model based on YOLOv8 produced expected results since it had high test accuracy of 97.75%, which was a marked improvement compared to the basic CNN (88.50%), the improved CNN with batch normalization (95.42%), and the ResNet50 transfer learning model (74.33%). The Residual connection, SiLU activators, and bottle neck blocks have facilitated the use of multi-scale features extraction and sturdy classification. All this proves that specialized CNNs architectures have the potential to offer consistent, automatic brain tumor recognition, which can be of great assistance to clinical decision-making and seem to enhance medical practice diagnoses.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Horn, Christian UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence R Medicine > Healthcare Industry 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:	24 Aug 2026 12:07
Last Modified:	24 Aug 2026 12:07
URI:	https://norma.ncirl.ie/id/eprint/9602

Actions (login required)

View Item