Enhanced Needle Visualization using Ultrasonic Synthetic Aperture Imaging with Specular CNN Modeling

Percutaneous needle interventions are extensively utilized for a broad range of medical procedures and therapies. Ultrasound imaging has been routinely used for guidance of the needle intervention due to its real-time capability and cost-efficiency. However, conventional ultrasound with fixed-angle transmission often suffers from reduced visualization of oblique needles when the specular echoes are reflected away from the receiving transducer. Synthetic Transmit Aperture (STA) beamforming with Delay and Standard Deviation (DASD) relies on the variation of echo magnitude among various incident angles to construct the image and thus improves the image contrast in needle detection. In this study, it is proposed to further integrate the specular information of the needle shaft with STA-DASD beamforming for enhancement of needle visualization. For comparison with the specular beamforming, the novel SpecularNet is also developed based on two efficient Convolutional Neural Networks (CNN) models to predict the specular information of image pixel. One model estimates the orientation of the needle from pixel-wise STA channel data and guides the transmit/receive apodization to emphasize the specular signal in STA-DASD image. The second model predicts the specular probability to suppress background noise and artifacts by image weighting. The proposed specular STA-DASD has been validated using simulations and experiments. Results indicate that, for the orientation of needle between 10° and 40°, the specular STA-DASD improves the needle-to-speckle ratio (NSR) by 15 dB and the needle-to-artifact ratio (NAR) by 25 dB compared to the conventional STA-DASD. Compared to specular beamforming, the inference to specular information can be performed 600 times faster in SpecularNet while the both NSR and NAR of the corresponding specular STA-DASD also improve by another 25-30 dB in the phantom experiments.