Bone age comparison using convolutional neural network

Abstract

In the last few years,Machine Learning has taken the world by storm. From predictive web browsing to the email and text classi cation,from the autonomous car to facial recognition, machine learning is the main core of every intelligent application that we can see now a days. Predicting bone age is another eld that has been bene ted exceedingly from the exposure of this technology. For this reason, we have proposed convolutional neural network for predicting the age of a child and doing a comparative analysis on with other available techniques. We have choose four models for it and they are: InceptionV3, VGG16, ResNet50 and MobileNet. By pre-processing the image and selecting the various parameters the framework has been trained and tested in "RSNA Pediatric Bone Age Machine Learning Challenge" dataset. Highest accuracy of 91.13% has been achieved for MobileNet with mean absolute error of 8.87, the explained variance score for this method is 0.92 and value loss during the training is 0.0809 whereas the lowest accuracy has been achieved for VGG16 with mean absolute error 32.58,the explained variance score for this method is 0.032 and value loss during the training is 1.0281.