Segmentación semántica 2D de la glándula prostática en imágenes de resonancia magnética usando redes neuronales convolucionales.

Abstract

Convolutional Neural Network (CNNs) is one of the most commonly used methods for automatic prostate segmentation in Magnetic Resonance Images (MRI). However, the results of the literature review in this work show that there is one research on the segmentation of the two main zones of the prostate gland (peripheral and central zones). This work proposed two different 2D semantic segmentation models to address the segmentation of the prostate in its main zones. The first model (Model-A) used an encoder-decoder architecture composed by a global and local U-net. The former segments the whole prostate, whereas the latter segments the central zone. The second model (Model-B) used an encoder-classifier architecture, where pixel by pixel classification was used to achieve the same segmentation goal. Moreover, model B used a VGGnet that was fine tuned. The Prostate-3T collection of the NCI-ISBI 2013 Challenge database was use to evaluate the performance of each model. The experimental results show a superior segmentation performance for Model A (DSC = 96.79% ± 0.15% and IoU = 93.79% ± 0.29%) compared to Model B (DSC = 92.50% ± 1.19% and IoU = 86.13% ± 2.02%)