Automatic Bone Age Assessment Combined with Transfer Learning and Support Vector Regression

Abstract

Bone Age Assessment (BAA) is an important topic in clinical practice of evaluating biological maturity of children. Traditional manually BAA is time consuming and with strong subjectivity. The existing automated BAA methods are based on pattern matching or classification methods, which are complex and with low generalization performance. To address these problems, a fully automatic BAA method combining transfer learning and regression learning is proposed. At first, the adaptive thresholding segmentation method is employed to segment hand bone from raw X-ray images. Then the transfer learning technique are used to extract high-level features of hand bone images. Finally, support vector regression is applied to perform BAA. The experimental results show that the proposed method is more accurately than the existing approaches and achieves better performance with lower root mean square error (RMSE)Bone Age Assessment (BAA) is an important topic in clinical practice of evaluating biological maturity of children. Traditional manually BAA is time consuming and with strong subjectivity. The existing automated BAA methods are based on pattern matching or classification methods, which are complex and with low generalization performance. To address these problems, a fully automatic BAA method combining transfer learning and regression learning is proposed. At first, the adaptive thresholding segmentation method is employed to segment hand bone from raw X-ray images. Then the transfer learning technique are used to extract high-level features of hand bone images. Finally, support vector regression is applied to perform BAA. The experimental results show that the proposed method is more accurately than the existing approaches and achieves better performance with lower root mean square error (RMSE).