Bone age estimation with HS-optimized Resnet and Yolo for child growth disorder

Abstract

Bone Age Assessment (BAA) plays a significant part in assessing skeletal maturity for preventing different types of skeletal disorders in the human body. Predicting BAA helps clinicians make decisions about the patient’s biological maturity. An increase or decrease in bone age (BA) causes serious issues for pediatricians who need advanced equipment for BA consistency with the chronological age. Recent techniques fail to accurately estimate the region over interest (ROI) of hand-wrist images to predict BA under low time complexity accurately. Hence, this article brings novel deep learning (DL) techniques to assess the BA to overcome child growth disorders accurately. The proposed methodology undergoes three major stages: pre-processing, ROI detection, and BA detection. In the pre-processing stage, contrast-limited fuzzy adaptive histogram equalization (CLFAHE) is done to maintain the originality of the hand image. The size of the image is also increased by using data augmentation. After pre-processing, the required hand bone region with ROI detection approaches you only live once version 3 (YOLOv3). At the final stage, the BA is predicted using the residual network-152 (ResNet152) technique. Finally, the weight updation takes place in the ResNet152 model using harmony search (HS) optimization techniques. The dataset used in this research is collected from the RSNA bone age dataset. The performance measures such as mean absolute error (MAE), root mean square error (RMSE), and root mean squared percentage error (RMSPE) are analyzed and compared with other existing measures. In an experimental scenario, the MAE of 2.59, RMSE of 2.97, and RMSPE of 0.35 for the male gender is obtained.