Computational Modelling and Machine Learning Based Image Processing in Spine Research

Abstract

Diagnosing spinal problems is not an easy task. Doctors collect different types of information,including magnetic resonance imaging (MRI),to make a final diag-nosis and decision on treatment modality. The localization of lumbar discs on MRI images is a challenging problem due to the wide range of variability in size,shape,number and appearance of discs and vertebrae. Current state-of-the art stud-ies show that most of the implemented methods are semi-automatic and suffer from additional correction of the solution or are very sensitive to the changes in parameters. This chapter will use two different approaches – computational model-ling using finite element method to investigate the displacements and stress dis-tribution and machine learning (ML) algorithms to perform automatic segmenta-tion of regions of interest (vertebrae,discs). The results for segmentation show high accuracy,with possibilities for improvement. Finite element analysis,per-formed on a 3-dimensional model automatically created from scans using ML,for a healthy and herniated disc,can provide an additional insight into the processes and different effect of the herniated disc onto the spine (i.e. back pain). A comput-er diagnostic system can be helpful in generating diagnostic results in short time and represent a help in final decision making.