Resumen

JIMENEZ-MURILLO, David et al. Brain Tissue Segmentation Using U-Net-based Convolutional Neural Networks. RISTI [online]. 2023, n.52, pp.5-24.  Epub 31-Dic-2023. ISSN 1646-9895.  https://doi.org/10.17013/risti.52.5-24.

Magnetic Resonance Imaging (MRI) is a non-invasive method that produces volumetric, high-resolution anatomy images. This method is commonly used for diagnosing various brain pathologies, such as demyelinating diseases, dementia, cerebrovascular diseases, and epilepsy, among others. However, manually segmenting tissues in these images is a complex task that requires a significant amount of time and effort from specialists. To address this, various strategies have been proposed in the literature that involve less expert involvement, both automatic and semi-automatic. In this study, a comparison is made between a convolutional neural network based on U-Net architecture and two other MRI segmentation methods, Dipy and FSL. The comparison uses three publicly available datasets, MRBrainS13, MRBrainS18, and IBSR, which contain MRI volumes with manually segmented brain tissues. The performance of each method is evaluated using four measures of similarity: The Dice coefficient, the Jaccard index, the area under the curve, and structural similarity. The results demonstrate that the methods based on convolutional neural networks achieve segmentations that are more similar to those performed manually by an expert compared to the other methods.

Palabras clave : Brain segmentation; convolutional neural network; deconvolution; medical image segmentation; magnetic resonance imaging.

        · resumen en Español     · texto en Español     · Español ( pdf )