Deep Learning Empowered Automatic Bone Age Assessment

Skeletal bone age assessment (BAA) is a commonly employed clinical practice used for the diagnosis of endocrine and metabolic illness in child growth. BAA approach generally starts with the acquisition of the X ray image of the left hand from the wrist to fingertip. The bones in the X ray image undergo comparison with the radiological images that exist in the standard atlas of bone development. Since manual methods are time consuming and erroneous, the recently developed deep learning (DL) models find useful in the design of automated BAA using X ray images. In this view, this paper presents a new DL empowered automated BAA (DL-ABAA) model using X ray images. The proposed DL-ABAA model performs initial preprocessing to improve the image quality. Followed by Fast region convolutional neural network (Fast-RCNN) with VGG-19 model-based feature extractor is involved for deriving the features from the input X ray images. At the same time, shuffled frog leaf optimization (SFLO) algorithm is utilized as a hyperparameter optimizer of the VGG-19 model. In addition, softmax (SM) based age prediction and extreme gradient boosting (XGBoost) based stage classification processes are applied to predict the age and determine the class labels. A detailed experimental results analysis stated the improved performance of the BAA technique over the recent approaches with the higher accuracy of 96.53%.