Multi Head Network for Effective Multiple Image Processing in the Medical Field

Abstract

Recently, with the development of convolutional neural networks, medical imaging studies are attracting attention. However, existing neural networks do not carefully consider that medical professionals perform their diagnosis by considering various anatomical structures. Therefore, we proposed a multi-head convolutional neural network to process multiple two-dimensional images effectively. Specifically, the model presented in this study firmly learns the unique characteristics of all input feature images effeiciently by using multi-head design and recent neural network techniques. As a result, our model achieves higher predictive performance with a smaller parameter size than the existing architectures for thyroid-associated ophthalmopathy patients' images acquired in an anonymous hospital. © 2022 IEEE.