Comparative analysis of GAN architectures for brain tumor segmentation in MRI images

Volodymyr I. Lytvynenko, Andrzej Smolarz, Paweł Komada, Mariia A. Voronenko, Saule S. Smailova, Volodymyr Manzhula, Victor M. Sineglazov, Kyrylo Lesohorskyi, Piotr Przybylski, Natalia Savina

2025

Closed access Metadata (CSL-JSON)

DOI 10.1117/12.3086457

Data confidence 70 medium

How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

This study presents a comparative evaluation of generative adversarial networks (GANs) for brain tumour segmentation in magnetic resonance imaging (MRI). We assess five GAN architectures – Attention GAN, U-Net GAN, SegGAN, WGAN-GP, and CycleGAN – using standard metrics such as Dice coefficient, IoU, Precision, Recall, F1-score, and Hausdorff Distance. The experimental results demonstrate that Attention GAN achieves superior segmentation performance, particularly in delineating irregular tumour boundaries. In contrast, CycleGAN exhibits limited accuracy due to its design for unpaired image translation. The findings highlight the importance of architecture choice and metric trade-offs in clinical applications, supporting the integration of GAN-based models into medical image analysis workflows.