Pipeline Design for Automated Vertebrae Segmentation for Paediatric Scoliosis Applications

Abstract

Scoliosis is abnormal lateral curvature of the spine exceeding the 10 degrees, being adolescent idiopathic scoliosis (AIS) the most prevalent type, identified in children between 10 and 16 years of age. Depending on the severity of the curvature, there are different treatment methods, with surgical treatment being the one of choice for cases with a curvature greater than 45 degrees. However, it presents numerous perioperative complications, therefore, the surgical planning of these cases is used to give support to the surgeon when performing the surgery. Efforts have been made to develop Artificial Intelligence (AI) based models for automatic vertebrae and spine segmentation, as this task is time-consuming and repetitive, and is used in numerous medical applications, including surgical planning. The aim of this project, started from scratch at the Hospital Sant Joan de Déu, is to design and develop a pipeline that integrates automatic segmentation methods to be used in AIS cases. Different automatic segmentation methods have been studied, among which one has been chosen and is used in the present project. Each of the steps to be carried out to implement the pipeline proposed with the method chosen are indicated, explained and detailed, including data collection, setup of the working environment, use of image pre-processing techniques, and the execution of the model. Additionally, this model has been re-trained with data from adolescents with scoliosis condition from the Hospital to improve its performance. Results show that the re-training of the model allows to obtain a better segmentation of the spine and vertebrae from the input CT images.