Samuel Kessler

Semester Work
Supervisors: Dr. Yuanhao Gong, Prof. Orcun Goksel

Estimating Elastograhy Inverse Problem using Simulation Data and CNN

The problem of estimating the elastic properties of cancer from elastography based on ultrasound images is demonstrated using simulation data and a convolutional neural network. This CNN performs well on simulated displacement fields from circular cancer shapes. The learned kernels are robust with parameter changes, such as shape parameters, vibration frequency, noise level etc. This method only contains 1665 parameters and thus can be used on mobile devices. Future work includes an extension to generalize the convolutional neural network for arbitrary shapes.