Pediatric Skeletal Age Assessment Using Deep Learning Proceedings

Skeletal age assessment (SAA) is one of the techniques to verify the claimed age and actual age. It is also performed to evaluate whether skeletal age is normal or delayed compared to the patient’s chronological age (CA). A delayed or advanced bone age can indicate growth disorders. It is generally performed by using either the Greulich & Pyle (G&P) method or the Tanner–Whitehouse (TW) method. However, inter- and intra-observer differences occur that could be resolved by developing an automatic system. Three years back (in 2017) Radiological Society of North America (RSNA) challenged the data science community to leverage computer vision to predict SA more efficiently and accurately than existing methods. The primary goal of this research work is to use convolutional neural networks to build a model to predict skeletal age with high accuracy given RSNA’s provided dataset. The secondary goal is to leverage deep learning visualization techniques for better interpretation of our results. Overall, our proposed model achieved a competitive MAE of 7.61 months on the test set provided by Radiological Society of North America (RSNA).