Bone Age Estimation with X-ray Images Based on EfficientNet Pre-training Model

Abstract

Human bones have different characteristics in different development stages, so the estimation of bone age can reflect the growth and development level of individuals relatively accurately. Bone age estimation aims to predict the biological age of children, which plays an important role on the diagnosis of some pediatric endocrine diseases. Tradition methods are carried out by doctors, and it is not effective in accuracy and speed. To this end, we proposed a deep-learning based method for bone age estimation. Based on the training set of more than 10000 X-ray images of hand bones from Radiological Society of North America (RSNA), this paper studies the processing, segmentation, feature extraction of X-ray hand bone images by using computer image processing and artificial intelligence learning methods, and uses convolution neural network to process samples and analyze them automatically. The main research work and achievements are as follows: (1) Pre-processing of X-ray hand bone image, unifying the size and cutting, reducing the image area without hand bone; (2) The gray-scale image is transformed into a three-channel image, and pre-processing by EfficientNet of ImageNet. Then convolution neural network is used to learn the features of X-ray hand bone image and evaluate it automatically. Finally, the network is evaluated by the minimum mean square error, so that the minimum mean square error is as close as possible to the minimum value. Through the neural network, the bone age from X-ray hand bone image can be quickly judged, and then it can be applied to clinical research.