Deep Learning for Automatic Determination of Bone Age in Children

Abstract

Skeletal bone age assessment (SBAA) is very important for both sides of parents and physicians to evaluate the irregular growth of children. SBAA process is carried out by radiologists who visually inspect the radiology image of the left hand according to the Greulich and Pyle (GP) or the Tanner-Whitehouse 2 (TW2) methods. However, human eyes have their own limitations and therefore the visual inspection procedure by radiologist involves a degree of error and also intra personal variability. To address these drawbacks, a deep learning-based approach is proposed here to precisely act on these X-ray images. The employed database contains 1391 X-ray left-hand image from Los Angeles children's hospital and 200 left hand x-ray image from different age and gender from Iranian children from Namazi hospital of Shiraz. Our results demonstrate the efficiency of proposed model (mean absolute error of 0.89) in this field.