Support vector machine classification using correlation prototypes for bone age assessment

Abstract

Bone age assessment (BAA) on hand radiographs is a frequent and time consuming task in radiology. Our method for automatic BAA is done in several steps: (i) extract of 14 epiphyseal regions from the radiographs, (ii) for each region, retain image features using the IRMA framework, (iii) use these features to build a classifier model, (iv) classify unknown hand images. In this paper, we combine a support vector machine (SVM) with cross-correlation to a prototype image for each class. These prototypes are obtained choosing the most similar image in each class according to mean cross-correlation. Comparing SVM with k nearest neighbor (kNN) classification, a systematic evaluation is presented using 1,097 images of 30 diagnostic classes. Mean error in age prediction is reduced from 1.0 to 0.9 years for 5-NN and SVM, respectively. © Springer-Verlag Berlin Heidelberg 2012.