Sequential CNN-Based Enhancement of Ultrafast Ultrasound Imaging for Sparse Arrays

With the emergence of ultra-portable ultrasound systems, the need to reduce cost and complexity of ultrasound systems has intensified. Sparse arrays, employing fewer trans-ducer elements to acquire images, offer a viable solution by effectively minimizing data and computational demands. However, in ultrafast ultrasound imaging, a reduced number of transducer elements results in increased artifacts, degrading image quality. This study proposes a convolutional neural network (CNN)-based approach to enhance image quality in ultrafast ultrasound using sparse arrays. While all transducer elements in the ultrasound probe are used to transmit unfocused wavefronts, only a sub-set is active during reception of echoes. The method employs sequential CNNs to refine image quality progressively. Each CNN in the sequence is trained to map images acquired using fewer transducer elements into images acquired with more active elements in reception. The final network is trained to transform single-plane-wave (PW) acquisitions to images resulting from the compounding of 87 PWs, acquired with all transducer elements. The sequential approach substantially reduces image artifacts, improving peak signal-to-noise ratio and structural similarity index measures compared to a single CNN method. These image quality improvements are particularly beneficial for portable and wireless ultrasound systems, where reducing data volume is an important incentive.