Homodyned-K quantitative ultrasound and machine learning for detection of lateral epicondylosis of the elbow
(2019) 2019 IEEE International Ultrasonics Symposium (IUS) — Location: Glasgow, United Kingdom (6.October.2019)
Files
Homodyned-K_quantitative_ultrasound_and_machine_learning_for_detection_of_lateral_epicondylosis_of_the_elbow.pdf
Open Access - Adobe PDF
- 418.21 KB
Details
- Authors
- Abstract
- Lateral epicondylosis (LE) of the elbow is a common syndrome found among working-age individuals, leading to degenerative changes of the common extensor tendon (CET). Ultrasound (US) is well suited for the investigation of LE because of its relative affordability and good spatial resolution. Among quantitative ultrasound (QUS) imaging techniques, homodyned-K (HK) statistical modeling of the echo envelope aims at characterizing tissue microstructures. The goal of this study was to assess the potential of HK parameters in detecting LE. In this prospective study, 30 LE elbows in 27 patients and 24 asymptomatic elbows in 13 volunteers underwent US imaging of the CET and radial collateral ligament (RCL). After US imaging examination per clinical standard practice, a long-axis, 3-second loop of a radiofrequency US image sequence of the CET and RCL was acquired using a Terason t3000 US scanner (Teratech, Burlington, MA) equipped with a linear 12L5-MHz transducer. Three statistical parameters based on HK modeling were estimated on the CET region-of-interest: 1) mean intensity normalized by its maximum value; 2) reciprocal 1/α of the scatterer clustering parameter; 3) coherent-to-diffuse signal ratio k. Moreover, HK parametric maps were calculated on the CETRCL region based on local estimation of the same parameters, from which were extracted additional features, as well as area of the two regions. Random forest classifier modeling identified the most discriminating combination of 3 features or less. The best combination of features was: CET global estimate of 1/α, CETRCL area, and inter-quartile range of local estimate of k. The area under the receiver operating characteristic curve, sensitivity, and specificity of the QUS-based model were 0.82 (95% confidence interval [CI], 0.80–0.85), 0.73, and 0.79, respectively. These values are comparable with values obtained in a meta-analysis: pooled sensitivity of 0.82 (95% CI, 0.76–0.87) and pooled specificity of 0.66 (95% CI, 0.60–0.72) when using US in the diagnosis of suspected LE.
- Affiliations
Citations
Bureau, N., Destrempes, F., Acid, S., Lungu, E., Moser, T., Michaud, J., & Cloutier, G. (2019). Homodyned-K quantitative ultrasound and machine learning for detection of lateral epicondylosis of the elbow. 2019 IEEE International Ultrasonics Symposium (IUS), Glasgow, United Kingdom. https://hdl.handle.net/2078.5/105814