Abstract
This paper describes a fully automatic computerized bone age assessment procedure based on phalangeal ossification features. Once a hand radiograph is read in, the bone age will be judged without any manual assistance. In the presented procedure, first a preprocessing stage, including finding the position of the left hand, defining and detecting phalangeal bone region of interest (PROI), and segmentation of the phalange by using Gabor filter, Canny edge detector, and local variance edge detector, is performed. Then two different sets of features, physiological and morphological features, are extracted from the segmented images to make bone age assessments using several kinds of neural networks. The effectiveness of the two sets of features is analyzed and the correct rates of the two sets are computed and compared. It is concluded that the morphological features perform better than the physiological features due to its ability to describe the development of bones. The best correct rates of assessed bone ages within 0.5 years errors, 1.0 year errors, 1.5 years errors, and 2.0 years errors, are 49.21%, 71.91%, 86.29%, and 93.25% in female, and 42.74%, 67.81%, 84.17%, and 91.02% in male, by virtue of leave-one-out cross validation, respectively. In a clinical aspect, errors within 1.5 years are acceptable and thus the system is satisfactory.