@article{article_1099510, title={iSeg-WNet: Volumetric Segmentation of Infant Brain MRI Images}, journal={Erciyes Üniversitesi Fen Bilimleri Enstitüsü Fen Bilimleri Dergisi}, volume={38}, pages={508–518}, year={2022}, author={Çelik, Gaffari}, keywords={Derin Öğrenme, CNN, Bölütleme, Dice Loss, 3D MRI Bölütleme, iseg-2019, iseg-2017}, abstract={Examination of infant brain development is extremely important in terms of early diagnosis of possible brain dysfunctions. Brain MRIs are examined by segmentation of white matter (WM), gray matter (GM) and cerebrospinal fluid (CSF) tissues. Low-density contrast between tissues in infant brains complicates the segmentation process. It is seen that the segmentation process is done very well with the Deep Learning architectures that have been developed recently. In this study, an architecture called Deep Learning-based iSeg-WNet is proposed for segmentation of infant brain MRI images. Appropriate hyperparameters were determined by different studies and the performances of different architectures were compared. Performance comparison was made according to Dice metric. In experimental studies, it has been observed that the use of MRI images in T1w and T2w images together increases the segmentation performance. At the same time, high performance was obtained by using Dice Loss as a cost function and MinMax normalization as a data normalization process. When the segmentation performances of different architectures are examined, it is seen that the proposed architecture segments CSF, GM and WM textures with the highest success. The proposed architecture is available at https://github.com/GaffariCelik/iSeg-WNet.}, number={3}, publisher={Erciyes Üniversitesi}