Introduction: Blood vessel diameter is an important metric in evaluating vascular function, whether it be assessing endothelial health through flow-mediated dilation, monitoring blood vessel responses to stimuli, or assessing general vascular health. To non-invasively collect these measurements, B-mode ultrasound footage is often recorded, with the process of measurements automated to ensure dynamic vessel wall movements from pulsatile flow are correctly accommodated for. However, current methods are often highly susceptible to intrinsic motion artifacts and visual noise due to the nature of B-mode recordings. Here, we introduce BWave.US, a new open-source program written in Python, offering an alternative for accurate dynamic blood vessel diameter measurements. Key improvements include a reduced the number of steps requiring user-input to maintain reproducible results and open-source nature which facilitates easy integration with other coding languages and software.
Materials and
Methods: Data for software refinement was attained from a previous study of 38 individuals. Individuals were recruited for this study who suffered a traumatic spinal cord injury at least 1 year prior, were ages 18-50, had their injuries classified using the American Spinal Injury Association Impairment Scale (AIS) and with neurological level of injury at or above T6. Age and gender matched peers without spinal cord injury were further recruited as a targeted control group. B-mode recordings were taken at the brachial artery at a point 5cm proximal to the antecubital fossa, and at an insonation angle of < 60degrees, over a recording of 15 seconds to measure baseline diameter. After baseline measures, a forearm cuff was placed 1cm distal to the antecubital fossa and inflated to vascular occlusion for 5 minutes to assess subsequent endothelial health through flow mediated dilation. Post-occlusion, B-mode recordings were again collected for three minutes to record the brachial artery diameter. These dynamic videos were selected to assess software refinements in a clinically representational and challenging environment. Comparisons between the current gold standard (FloWave.US in MATLAB) and our updated code (BWave.US) were done by modifying the FloWave.US BMode.m script to use the same region of interest (ROI) for each video during head-to-head comparison. Objective comparisons were made between programs with regard to frame capture rate, reproducibility, and diameter measurements. A Bland-Altman plot was generated to assess for measurement biases.
Results, Conclusions, and Discussions: After identifying arterial wall detection deficits with FloWave.US, BWave.US was created. BWave.US first preprocess frames with equalized contrast and Gaussian blur. Sobel edge detection and Otsu Binary thresholding then inform Watershed image segmentation. Contacting pixels are grouped contiguously, and two vessel edges are identified from pixel groups, fitting two parallel lines by minimizing least-squares regression. Diameter is calculated from the distance between the two parallel lines. After analyzing each frame, signals are filtered with high-pass and standard deviation bands to remove noise in the physiological data. We analyzed 68 total videos in both FloWave.US and BWave.US. Per video, FloWave.US was able to read a median of 100 (98.0-100) % of all frames in the baseline condition, whereas BWave.US read 89.1 (83.0–93.7)%. In the post-occlusion setting, FloWave.US read 93.4 (65.7-100) % of all frames, whereas BWave.US read 85.8 (79.5–89.9) % of all frames. Overall, FloWave.US read 99.4 (82.9-100) % of all total frames, while BWave.US read 87.6 (81.4-92.7) %. Fl oWave.US, the current gold standard, preserved more readable frames under baseline conditions where there was minimal motion artifact. However, in the post-occlusion condition , which created motion artifacts due to clinically important dilation of blood vessels, BWave.US outperformed FloWave.US. Additionally, the higher interquartile range in the percentage of frames read per video demonstrates the higher variability in FloWave.US in this post-occlusion setting. This critically improved complete usable dataset measurements from 55% from FloWave.US to 89% with BWave.US. In contrast to FloWave.US, the ROI used in data collection is saved along with the diameter recordings, enabling users to replicate results consistently. Additionally, BWave.US saves diameter measurements in their raw, unfiltered form, unlike FloWave.US , which only saves post-processed data. This gives users the flexibility to apply their own data filtering methods. Furthermore, our Bland-Altman analysis demonstrated that BWave.US had a fixed positive 4.43-pixel bias in diameter readings (~0.35mm) when compared to FloWave.US (R2=0.282, p< 0.00001). FloWave.US proves to be useful when there is little-to-no expectation for change in blood vessel diameter, but BWave.US may provide an improved alternative when more motion artifact is present in the video or vessel dilation is expected.
Acknowledgements (Optional): We would like to thank the Assistive Restorative Technology Laboratory at the Mayo Clinic.
