Implementation of a deep learning model for vertebral segmentation in CT data

Abstract

This paper deals with the problem of vertebral segmentationin CT data with the use of deep learning approaches.Automatic segmentation of vertebrae is a very complex issueand would simplify the work of radiologists and doctors. Thepaper is focused on one of the models published and submittedto the Large Scale Vertebrae Segmentation Challenge (VerSe) in2020 from C. Payer et al. – Improving Coarse to Fine VertebraeLocalisation and Segmentation with SpatialConfiguration-Netand U-Net and its implementation and modification. The modelis evaluated on the corresponding public and hidden dataset. Itsmodification shows an improvement of the results in comparisonwith the published results, a mean Dice score improved from0.9165 to 0.9302 on the public dataset and from 0.8971 to 0.9264on the hidden dataset.