Machine learning assisted filtering of Focused Ultrasound pulse-induced interference in Harmonic Motion Imaging (HMI) derived displacement

Harmonic Motion Imaging (HMI) is an acoustic radiation-force (RF) based elasticity imaging technique which can be used to monitor focused ultrasound (FUS) induced ablation without interruption. Our group has previously demonstrated frequency domain filtering in removing FUS interference from imaging data. However, FUS interference becomes more pronounced at increased intensities and durations, and various imaging and tracking transducer configurations may cause an overlap in frequency spectra between the FUS and imaging pulse, making frequency domain filtering unsuitable. FUS-net, a CNN-based UNET autoencoder developed by our group, was previously shown to be able to reduce FUS artifacts in RF data acquired with single acoustic radiation force (ARF) pulses with a duration of 1 to 2 ms. In this study, FUS-net was applied to reduce FUS artifacts in an RF dataset acquired during HMI, with continuous wave oscillating pulses for a duration of 6s. The FUS-net model was trained and tested on beamformed RF data acquired from an in vivo breast cancer mouse model (N=3) and ex vivo human fibroadenoma breast tissue (N=1). Notch filtered beamformed RF data was used as ground truth. Displacement estimated from FUS-net processed data was compared to displacement estimated from notch filtered RF data. We showed less noise in the focal area of HMI displacement maps in the FUS-net filtered dataset as compared with frequency-based filtering, with a median coefficient of variation of 0.7 and 0.2 in displacement calculated from notch filtered RF data and FUS-net filtered data respectively.