Combined manual and automatic landmark detection for enhanced surface registration of anatomical structures: an extensive parameter study for femur and clavicle

Abstract

ABSTRACTSurface registration, in which a mapping between surfaces is calculated, is a powerful tool to create patient-specific musculoskeletal and statistical shape models. However, to create these models, surface registration must assure a one-to-one correspondence between both surfaces. To enhance this process, a surface registration framework that uses a combination of manual and automatic landmark detection is presented in this study. In addition, an extensive parameter study of the framework for a femur and a clavicle is conducted. The average correspondence quality of nine femur landmarks decreased from 15.6 mm using automatic point detection to 6.1 mm by combining manually and automatically indicated landmarks. For the clavicle, the average distance of five landmarks decreased from 2.8 to 0.9 mm. Combining manual and automatic landmark detection clearly improved correspondence quality. Results confirmed the applicability of the proposed registration framework for femur and clavicle, although with different parameter settings.