A study of bone age evaluation based on hand knuckles radiogram

Abstract

The aim of this paper is to evaluate children's bone age based on the radiogram of epiphyseal/metaphyseal region of interests (EMROIs). At first, we take the 9 EMROIs from ring finger, middle finger and index finger into our analysis and then extract the 13 geometrical features for each of them. Next, we utilized the KNN classification under the binary decision tree structure for determining the bone age. In the classification phase, two algorithms were considered. For the first one, features from the 9 knuckles were concatenated into one before the classification. The second one is to analyze the features of the individual 9 knuckles to produce 9 bone age results, and then to choose their mode as the final bone age. Our experiments demonstrate the proposed algorithm approaches the accuracies of about 60% in 1 year error and 80% in 2 year error for the first algorithm, and the accuracies of about 65% in 1 year error and 80% in 2 year error for the second algorithm, respectively. The experimental results reveal that the proposed bone age evaluation can provide a reliable and effective assessment for aiding the pediatricians in clinic.