Real-Time Processing of 2D and 3D Ultrasound Localisation Microscopy: from Radiofrequency to Super-Resolution

Quantifying the functional and morphological microvascular details of tumours and tissue could potentially improve early detection and classification of vascular diseases. Super-Resolution Ultrasound (SRUS) visualises vascularisation and flow with microscopic resolution, superior to what is currently clinically available. SRUS achieves micrometre resolution by localising microbubbles (MBs) travelling in the microvasculature. Despite the potential of SRUS, its long processing time has limited its clinical translation. This work presents a graphics processing unit (GPU) processing pipeline that generates 2D and 3D SRUS images faster than their imaging frame rate, thereby achieving real-time processing performance. The processing pipeline includes radiofrequency data transfer from random-access memory (RAM) to GPU, beamforming, envelope detection, background removal, spatial filtering, MB isolation and localisation, SRUS image generation, and SRUS results transfer from GPU to RAM. The processing pipeline was implemented using C/C++ and CUDA. The presented processing pipeline computed 2D and 3D SRUS data at a rate of 500 and 100 compounded frames/volumes per second, respectively, using consumer-grade GPUs (<$1000).