Convolution neural network (CNN) is a state-of-the-art method that is widely used in the field of image processing. However, one major limitation of CNN is that it does not consider the spatial orientation of the image. Capsule network, proposed by Geoffrey E. Hinton et al., was an attempt to solve this limitation. However, the architecture was designed for discrete data. This paper modifies the architecture appropriately to make it suitable to work on continuous data. It works on the dataset RSNA Pediatric Bone Age Challenge (2017) (RSNA Pediatric Bone Age Challenge in Stanford medicine, Dataset from https://www.kaggle.com/kmader/rsna-bone-age (2017) [1]) to detect the bone age of the patient from his X-ray, whose maximum age is restricted to 228 months. In order to achieve the purpose, mean squared error (MSE) was used for backpropagation. The 20 most significant outputs were taken from the network to address the problem of diminishing gradients. The results were validated to check if it is biased to an age range. This could be a characteristic for running on continuous data using an architecture that supports the classification of only discrete data. Since the validation held true, one could infer that this network could be more suitable for continuous data than capsule network.
CITATION STYLE
Koppar, A., Kailasam, S., Varun, M., & Hiremath, I. (2021). Pediatric Bone Age Detection Using Capsule Network. In Lecture Notes in Networks and Systems (Vol. 173 LNNS, pp. 405–420). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-33-4305-4_31
Mendeley helps you to discover research relevant for your work.