A Deep Learning Signal-Based Approach to Fast Harmonic Imaging

High resultant image contrast and quality have caused tissue harmonic imaging to become a valuable tool in ultrasound imaging. Amplitude Modulation (AM) is one of the most commonly used nonlinear pulsing schemes in tissue harmonic imaging. However, its need for at least two consecutive firings continues to be a hindering factor for a faster imaging process. In this work, deep learning concepts are exploited to introduce an alternative approach for ultrasound tissue harmonic imaging using a single firing. This is achieved by implementing an asymmetric convolutional autoencoder network to estimate the low-harmonic component content from a received echo signal. The network is trained on the full-amplitude harmonic content IQ echo comprising the network's input and its corresponding low-amplitude harmonic IQ echo representing the network's output. The proposed approach yielded high contrast harmonic images with comparable contrast to noise ratio and contrast ratio to the conventional checkerboard apertures amplitude modulation technique, yet at approximately three times the frame rate. Moreover, less clutter is observed in the proposed approach reconstructed images in contrast with the ground truth images. These results open the door for the implementation of harmonic imaging with comparable quality to the conventional AM techniques, yet with an increased framerate and reduced motion artifacts.