Capsule Neural Network can encode position and pose information of objects by its dynamic routing algorithm. This local search algorithm computes coupling coefficients. The Capsule network has been tested on regular and medical datasets with classification and segmentation tasks. Results show it outperforms regular convolution neural networks with fewer parameters. In recent years, research focuses on improving its routing algorithm generally. However, it also has disadvantages such as low training speed. It has not been tested on a complex dataset either. In this presentation, we will first introduce the vanilla capsule network and then talk about its applications in medical imaging.

Ultrasound is a great widely used method for medical diagnostic imaging. It has the advantage of being able to work in real-time and can be easy to operate. However, due to the physical limitations of Ultrasound including wavelengths, attenuation in tissue, and trades offs between resolution and tissue penetration; clear imaging and deep microscopy have been significant hurdles to overcome. New techniques such as super-resolution imaging using micro-bubbles and the utilizing of convolutional neural networks (CNNs) have been used to be able to overcome these challenges to provide better diagnostic imaging.