Thapa, Samima (2026) Image segmentation using Deep Learning for Brain Tumor Segmentation. Masters thesis, York St John University.

Text (MSc by Research thesis) Image segmentation using Deep Learning for Brain Tumor Segmentation.pdf - Published Version Restricted to Repository staff only until 31 July 2026. Available under License Creative Commons Attribution. Download (3MB)

Abstract

Brain tumours are among the most life-threatening diseases, and early detection plays a crucial role in improving patient survival. Accurate brain tumour segmentation is essential for diagnosis, treatment planning, and monitoring tumour progression. Manual segmentation is time-consuming
and subject to inter-observer variability, whereas automated segmentation methods address these limitations by improving efficiency and consistency. This study focuses on automated brain tumour segmentation using a deep learning–based U-Net model trained on the BraTS 2020 dataset. The dataset includes multimodal MRI scans (T1, T1Gd, T2, and FLAIR) along with expert annotations of tumour core, enhancing tumour, and edema regions. A convolutional neural network (CNN) was first implemented as a baseline model, followed by training and fine-tuning of the U-Net model to achieve optimal performance. The proposed U-Net model, evaluated on
the BraTS 2020 dataset, achieves competitive Dice scores of 0.8852, 0.8739, and 0.8597 for WT, TC, and ET, respectively. Dataset imbalance was identified as a major challenge affecting model accuracy and was addressed by combining weighted cross-entropy and Dice loss functions. Despite these improvements, accurately identifying the boundaries of small tumour core regions remained difficult, representing a key limitation of the model and an area for future research.

Item Type:	Thesis (Masters)
Status:	Published
Subjects:	T Technology > T Technology (General)
School/Department:	London Campus
URI:	https://ray.yorksj.ac.uk/id/eprint/14735

University Staff: Request a correction | RaY Editors: Update this record