B C Fan1, C W Hsieh, T L Jong, C M Tiu. 1. Department of Electrical Engineering, National Tsing-Hua University, Hsinch, R.O.C.
Abstract
BACKGROUND: Bone age (BA) estimation is one of the important applications of hand radiography in the area of pediatrics, especially for the diagnosis of endocrinological problems and growth disorders. BA estimation (BAE) is tedious, time-consuming and highly dependent upon the expert experience. Nowadays, most BAE standards are based on American standards, but there is no BAE standard for Taiwanese people. We attempt to construct a computerized BAE system to automate BAE and in the long run to build a BAE standard for Taiwanese. METHODS: Our BAE system is based on the carpal bone information. We propose a new 2-stage edge detection method for carpal bone feature extraction and a new method for locating the carpal bone region of interest. RESULTS: After the image is manually equalized, our BAE system can estimate the bone age automatically. The extracted carpal bone features were applied to three classifiers for age estimation: the weighted minimum distance, Bayes, and neural network classifiers. The Bayes and neural network classifiers had better results. In the 0.5-year tolerance case, they both had over 90% correct rate for both male and female training data. In the 1-year tolerance case, they could classify correctly for the male and female testing data. CONCLUSIONS: A computerized BAE system has been developed and some experiments have been conducted. It is found that the classifying results for 0.5-year tolerance are good and satisfactory.
BACKGROUND: Bone age (BA) estimation is one of the important applications of hand radiography in the area of pediatrics, especially for the diagnosis of endocrinological problems and growth disorders. BA estimation (BAE) is tedious, time-consuming and highly dependent upon the expert experience. Nowadays, most BAE standards are based on American standards, but there is no BAE standard for Taiwanese people. We attempt to construct a computerized BAE system to automate BAE and in the long run to build a BAE standard for Taiwanese. METHODS: Our BAE system is based on the carpal bone information. We propose a new 2-stage edge detection method for carpal bone feature extraction and a new method for locating the carpal bone region of interest. RESULTS: After the image is manually equalized, our BAE system can estimate the bone age automatically. The extracted carpal bone features were applied to three classifiers for age estimation: the weighted minimum distance, Bayes, and neural network classifiers. The Bayes and neural network classifiers had better results. In the 0.5-year tolerance case, they both had over 90% correct rate for both male and female training data. In the 1-year tolerance case, they could classify correctly for the male and female testing data. CONCLUSIONS: A computerized BAE system has been developed and some experiments have been conducted. It is found that the classifying results for 0.5-year tolerance are good and satisfactory.