Automated segmentation of epiphysis and metaphysis for computerized bone age analysis

Bone age estimation from hand wrist radiographs is a standard procedure for assessing skeletal maturity in pediatric radiology. This is needed for monitoring growth disorders due to disease or therapeutics, or establishing birth details. The estimation methods are based on analysis of ossification of the epiphysis of distal, middle and proximal phalanx of the left hand. Currently the analysis is done manually through subjective evaluation by raters. Even trained raters have considerable variability in assessment, primarily due to difficulty in estimating and comparing perceived ossification visually, and secondly, due to lack of time. Automated segmentation of epiphysis and metaphysis is crucial in automating the procedure, which apart from saving radiologists' time, would facilitate repeatability and reproducibility of analysis. In this paper, we describe a completely automated method for extraction of epiphysis. Segmentation of the epiphysis and metaphysis is done in the following steps: pre-processing for intensity adjustments, hand orientation correction, finger alignment, extraction of fingers, extraction of phalanges, and extraction of epiphysis from the phalanges. The algorithm has been validated on a digital atlas data with near 95% accuracy, and the failure modes are detection of absence of the epiphysis.