Model-based segmentation of pediatric and adult joints for orthopedic measurements in digital radiographs of the lower limbs

摘要

The growth of human bones forms a major problem when automatically segmenting orthopedic radiographs. Any template-based segmentation methods fails to fully capture these non-linear developments. However to extract orthopedic measurements or the bone age for patients of arbitrary age it is mandatory to have a segmentation scheme that deals with growth related changes. In this paper we propose a robust method based on Active Shape Models (ASMs) that on the one hand is invariant against the patient's age and on the other hand generalizes well over the large inter-patient variability. Our method achieves an accuracy of 0.48 mm for adult patients and 0.64 mm for children on a large test set of 180 images, with the patient's age covering a high range from less than one month to 93 years.