Statistical Shape Model of Costal Cartilage in the Human Body: Using Skeletonization, PCA and Regression to explore population variability

Abstract

Human Body Models are digital representations of the human anatomy used for crash test injury prediction. They provide an alternative to physical crash test dummies by providing more exact injury predictions while being cheaper to employ. Vehicle crashes are the most common cause of blunt thoracic injuries. The costal cartilage connects the ribs to the sternum and injuries to the costal cartilage is believed to cause ribcage instability, pain and prolonged disability. The geometry of the costal cartilage is sparsely studied. Not only does the shape and size of the costal cartilage vary, the topology differs between individuals. In this thesis, the topology of the costal cartilage is studied using Skeletoinzation and graph search algorithms. Using the most common topology, a template mesh is cast to the subject geometries using Coherent Point Drift and Iterative Closest Point. The populational variance is studied using Principal Component Analysis, resulting in a Statistical Shape Model of the costal cartilage. Finally, Linear Regression is used to correlate the variations to the demographic variables Age, Sex, Height and BMI. The resulting regression model obtains an R2 of 0.23, which is similar to the values obtained when studying equally complex geometries. The results indicate that the vertical spacing between cartilages increase with height, and that the width of the costal cartilage increases with BMI.