Super-Resolution with Embedded Denoising via Image Frequency Separation and Convolutional Neural Network in a Prototyped Transcranial Ultrasound Brain Imaging Scanner

Medical ultrasound (US) imaging represents a remarkable technological breakthrough, offering advantages such as portability, affordability, safety, rapid imaging capabilities, and a standard diagnostic quality. The successful development of under-skull transcranial US (TUS) imaging has the potential to revolutionize the field of quantitative ultrasound diagnostics, providing a competitive alternative to current expensive diagnostic modalities like CT scans and MRI. Despite the promising aspects, researchers face significant challenges in obtaining high-resolution images with adequate tissue contrast essential for precise diagnosis and treatment planning. To overcome these limitations, innovative techniques are being explored. Among the primary obstacles, achieving high-resolution imaging beneath the adult skull remains a critical hurdle due to the high acoustic impedance of the skull, as well as scattering and absorption of signals by brain tissues. In addition, extending the imaging depth without sacrificing image quality is another area of active research, closely tied to resolution and potential image quality loss. Furthermore, the sensitivity of TUS to patient movement results in motion artifacts that substantially degrade image quality. Addressing and compensating for these motion artifacts through intelligent and robust super-resolution (SR) based analysis persist as ongoing challenges. TRUBI (Transcranial Ultrasound Brain Imaging System) is an innovative 3D transcranial ultrasound brain imaging device developed by Tessonics^®. This cutting-edge system aims to overcome the primary limitations associated with conventional transcranial imaging, specifically the significant distorting effects caused by the human skull. Apart from the hardware design and interface software of the device in general, theoretical design, analysis and implementation of the signal and image processing units of the system is one of the main challenges. In this paper, we provide a concise overview of the convolutional neural networks (CNN) based simultaneous image SR and denoising, which holds the potential to be further developed to be incorporated into future generations of TUS imaging systems.