Design and development of vision transformer-based multi-scale MobileNet for skeletal bone age assessment with image features

Abstract

In recent days, skeletal bone age assessment gained more attention from researchers for auxiliary diagnosis and prediction in medical issues. Hence, it is essential to undertake more than a few complications existing in the traditional techniques in order to enhance the age assessment efficiency. So, this research work presented an intellectual skeletal bone age estimation method with neural network learning. At first, the needed images are gathered as of the available online source. Further, from gathered images, three sets of features are extracted from Vision Transformer (ViT) features as F1, Local Weber Pattern (LWP) image as F2, and raw image as F3. Subsequently, the resultant features are fed into the Vision Transformer-based Multi-scale MobileNet (ViT-MMNet) designed for assessing a skeletal bone age. In this network, multi-scale MobileNet is considered to execute the prediction process, where the ViT-based features are given to the first convolution layer, LWP-based features are given to the second convolution layer, and finally, in third convolution layer, the original images collected from the datasets are directly given. Finally, detailed experimental validations are conducted for the designed skeletal bone age assessment framework through traditional methods to assure the efficacy of the method.