SEGMENTATION OF BRAIN MRI IMAGES USING A DEEP LEARNING BASED U-NET ARCHITECTURE


Tayan E., Bozkurt F.

6th International Topkapi Congress on Current Developments in Scientific Research, İstanbul, Türkiye, 9 - 11 Ocak 2026, ss.199-206, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.199-206
  • Atatürk Üniversitesi Adresli: Evet

Özet

Early diagnosis of brain tumors is of great importance for preserving quality of life. The automatic segmentation of brain tumor regions in magnetic resonance imaging (MRI) enables faster clinical decision-making. In recent years, advancements in deep learning methods have made it possible to identify abnormal tissue regions more rapidly and effectively using MRI images.

In this study, the Brain MRI Segmentation Dataset, obtained from The Cancer Imaging Archive (TCIA) and openly available on the Kaggle platform, was used to investigate the segmentation of FLAIR abnormalities in brain MRI images. The dataset consists of brain magnetic resonance (MR) images acquired from 110 patients along with corresponding manually annotated FLAIR abnormality segmentation masks. The images used in the study were subjected to basic preprocessing steps and prepared as suitable inputs for the segmentation model.

The segmentation process was carried out using a U-Net–based deep learning architecture. The segmentation performance outputs were evaluated using numerical metrics based on manual reference masks. The findings obtained from the analysis indicate that the U-Net architecture is an effective method for the segmentation of FLAIR abnormalities. This study presents a reproducible methodology for automatic segmentation applications using MR images.