Abstract
Bone age assessment is one of the most important problems to assess Pediatric growth. Various methods have been investigated to develop bone age assessment. The study of bone age assessment helps to diagnose disorders in growth progress. It is very common to monitor growth progress with left hand x-ray images. There are many method such as Tanner-Whitehouse [TW], Greulich and Pyle (G&P) methods. In this paper, we applied convolutional neural network resnet50 by preprocessing the images using different filters based on edge detection to compare effects on CNN accuracy. The main purpose of the paper is to compare the effect of DoG filtering and à trous wavelet as preprocessing techniques on bone-age assessment. It has been proved that suggested methods improved the accuracy of CNN (resnet50) in filtered images compared to the result of the non-filtered ones. The ages between 0 and 7 are analyzed to assess bone for female, male, and female male together. It is observed that there is over% 14 improvement for Female for Male %11 and for male and female %10 percent between filtered images and non-filtered images. The proposed method reached% 87.78 female and% 80 for male within 7.9 month.