Analysis of Deep Learning Approaches for Cervical Spine Segmentation

In this paper, we analyse Deep Learning methods in medi- cal imaging domain. Further, we apply various Convolutional Neural Network (CNN) architectures that are commonly used in traditional computer vision for segmentation of 2D images to segment cervical spine in Magnetic Resonance Images (MRI). We compare performance of selected CNN models using established metric. Moreover, we investigate the ability of the CNNs to reuse existing knowledge as well as their capability to generalize on medical data. Additionally, we analyse the impact of randomly initialized weights and bias correction of the data on the performance of the models.