Author: Cristel Baiu, Zheng Feng Lu, Dufan Wu, Baihui Yu 👨🔬
Affiliation: University of Wisconsin, Massachusetts General Hospital, University of Chicago 🌍
Purpose: IEC TS 62791:2022 specified a clinically meaningful and quantitative framework for evaluating diagnostic ultrasound performance by measuring the lesion signal-to-noise ratio (LSNR) using randomly-distributed hypoechoic sphere phantoms and mechanical scanning devices. Building on this, we developed an automated analysis method tailored for freehand scanning. This study evaluates the robustness of our method by identifying sources of variability and analyzing their impact on LSNR. Our ultimate goal is to ensure reliability in detecting suboptimal imaging performance in clinical applications.
Methods: A 2mm spherical voids phantom was scanned by a GE Logiq E10 system with ML6-15 transducer at 15MHz. An operator scanned evenly freehand during a fixed 4.13s cine acquisition, covering distances of 2, 3, 4, 5, and 6cm at increasing scanning speeds. For the 6cm coverage, five operators conducted acquisitions. Ten repetitions of each acquisition were performed to assess repeatability. The analysis algorithm stacked 62 frames from cine into 3D volumes, segmented individual spheres using depth-adaptive thresholds, and calculated individual sphere’s LSNR compared to its surrounding background at the sphere’s maximum cross-section plane. LSNRs within the same depth slabs were averaged to generate LSNR-depth curve.
Results: Variability was quantified as the relative standard deviation averaged along depth for both LSNR and sphere count curves. For repeated scans, LSNR variability ranged from 2.5% to 6.9%, despite sphere count variability ranging from 9.1% to 25.5% across all settings. The scanning-speed-induced variability was calculated by first averaging ten curves under each scanning speed and then calculating variability. LSNR showed a small variability of 5.5%, compared to the large sphere count variability of 31.3%. Similarly, the inter-operator variability was calculated, the LSNR variability being 5.2% and the sphere count variability 8.5%.
Conclusion: The automated analysis algorithm for freehand scanning demonstrated robustness, achieving approximately 5% variability in LSNR across various scanning speeds and operators