Evaluation of the algorithm for automatic identification of the common carotid artery in ARTSENS

Abstract

Arterial compliance (AC) is an indicator of the risk of cardiovascular diseases (CVDs) and it is generally estimated by B-mode ultrasound investigation. The number of sonologists in low- and middle-income countries is very disproportionate to the extent of CVD. To bridge this gap we are developing an image-free CVD risk screening tool–arterial stiffness evaluation for non-invasive screening (ARTSENS™) which can be operated with minimal training. ARTSENS uses a single element ultrasound transducer to investigate the wall dynamics of the common carotid artery (CCA) and subsequently measure the AC. Identification of the proximal and distal walls of the CCA, in the ultrasound frames, is an important step in the process of the measurement of AC. The image-free nature of ARTSENS creates some unique issues which necessitate the development of a new algorithm that can automatically identify the CCA from a sequence of A-mode radio-frequency (RF) frames. We have earlier presented the concept and preliminary results for an algorithm that employed clues from the relative positions and temporal motion of CCA walls, for identifying the CCA and finding the approximate wall positions. In this paper, we present the detailed algorithm and its extensive evaluation based on simulation and clinical studies. The algorithm identified the wall position correctly in more than 90% of all simulated datasets where the signal-to-noise ratio was greater than 3 dB. The algorithm was then tested extensively on RF data obtained from the CCA of 30 human volunteers, where it successfully located the arterial walls in more than 70% of all measurements. The algorithm could successfully reject frames where the CCA was not present thus assisting the operator to place the probe correctly in the image-free system, ARTSENS. It was demonstrated that the algorithm can be used in real-time with few trade-offs which do not affect the accuracy of CCA identification. A new method for depth range selection that leads to significant performance improvements has also been demonstrated.