Bone Age Assessment By Deep Convolutional Neural Networks Combined With Clinical Tw3-Rus

Xiaohong Liu,Guangyu Wang,Yiming Xu,Ning Chen,Ting Chen

DOI: https://doi.org/10.1109/BIBM47256.2019.8983019

2019-01-01

Abstract:Bone age assessment is critical to diagnosis of various growth disorders in children, such as endocrine, nutritional disorders and dysplasia. X-rays of hand and wrist are the most common modality used to calculate bone age. In this paper, we propose a novel approach called DeepTW3 for automatic bone age assessment from X-ray images. DeepTW3 integrates Convolutional Neural Networks (CNNs) with expertise knowledge of TW3(Tanner-Whitehouse 3nd edition)-RUS(radius, ulna and short bones) bone age assessment system. The proposed method is tested on a dataset containing 1,100 hand bone X-ray images, all of which were manually annotated with selected region of interests(ROIs). Our method achieved mean absolute errors (MAE) of 0.2685, outperforming all state-of-the-art methods. For the task of grading skeletal maturity, our method using continuous stage distribution is complementary to using the clinical TW3-RUS categorical stages when interpreting critical cases of intermediate bone stage.