Login / Signup

Muscle thickness assessment of the forearm via ultrasonography: is experience level important?

Vickie WongJun Seob SongTakashi AbeRobert W SpitzYujiro YamadaZachary W BellRyo KataokaMinsoo KangJeremy P Loenneke
Published in: Biomedical physics & engineering express (2022)
It is suggested that experience is needed in order to capture valid estimates of muscle size with ultrasound. However, it is unknown whether there is a large degree of skill needed to analyze the images once they are captured. Objective. To determine if less experienced raters could accurately analyze ultrasound images of the forearm by comparing their estimates with those of a very experienced ultrasonographer (criterion). Approach. 50 muscle thickness images were captured by one experienced ultrasonographer (also Rater 1). Those images were saved and were then measured by four raters with different levels of experience. The rater who captured the images was very experienced (criterion), the second rater was also experienced and provided 5 minutes of instruction for Rater 3 (minimal experience) and Rater 4 (no experience). Test-retest reliability was also determined for Rater 3 and 4. Main Results. The average muscle thickness value for the criterion was 3.73 cm. The constant error for Rater 2, 3, and 4 was -0.003 (0.02) cm ( p = 0.362), -0.07 (0.04) cm ( p < 0.001), and 0.02 (0.09) cm ( p = 0.132), respectively. The SD for Rater 4 was greater, resulting in wider limits of agreement compared to Rater 2 and 3. Absolute error was 0.01 cm for Rater 2, whilst it was 0.07 cm and 0.03 cm for the two inexperienced raters (Rater 3 and 4). The error for Rater 3 was systematic and post-hoc assessment found that this rater was using a different border than the other three raters (but consistent across time). For the test-retest reliability, the minimal difference for Rater 3 was 0.08 cm (relative minimal difference of 2%) and 0.17 cm (relative minimal difference of 4%) for Rater 4. Significance. Less experienced raters were able to accurately and reliably analyze already captured muscle thickness images of the forearm with low absolute errors.
Keyphrases
  • optical coherence tomography
  • deep learning
  • convolutional neural network
  • skeletal muscle
  • magnetic resonance imaging
  • magnetic resonance
  • machine learning
  • computed tomography
  • patient safety
  • contrast enhanced