Join now to get access to this content and more.
Become a SOAP member and have access to our benefits.
- For Review: SOAP Consensus Statement on Neuraxial Procedures in Thrombocytopenic Parturients
- Sample Centers of Excellence Applications
- ASA Corner
- SOAP Policy and Procedure Manual (P&P Manual)
- SOAP Expert Opinions
- SOAP's Learning Modules
- 2019 Annual Meeting Lecture Videos
- December 2018 - SOAP Unofficial Guide to ASA Committees Webinar
- Submit a Position
- View Job Postings
- Previous Meeting Archives
- Previous Meeting Abstract Search
- CMS Guidelines
- Member Benefits
- Newsletter Clinical Articles
- ACOG Documents
- Search our Patient Safety Archive
- Ask SOAP a Question
- Global Health Opportunities
- And more…
AUTOMATIC DETECTION OF THE LIGAMENTUM FLAVUM ON ULTRASOUND IMAGES OF THE LUMBAR SPINE
Abstract Number: 24
Abstract Type: Original Research
Introduction: Ultrasound has proven beneficial for guiding needle insertions in regional anesthesia  and more recently in lumbar epidural anesthesia . For epidural needle placement, the ligamentum flavum (LF) is the primary image landmark but it can be challenging to detect without ultrasound scanning experience. A new computer algorithm is developed and tested for automatically locating the LF depth on ultrasound images.
Methods: 22 parturient subjects participated with signed consent and review-board approval. The algorithm extracted a map of the bones (lamina) and LF from the image using an extension of phase symmetry analysis . The lamina and LF were located using a template-matching method. The algorithm was tested on 2 lumbar intervertebral spaces per subject, for a total of 43 images (one subject had only one image). This depth of the LF was compared to the depth measured (1) by manual segmentation, (2) manually by a sonographer, and (3) the needle insertion depth from an anesthesiologist performing loss-of-resistance.
Results: The algorithm successfully identified the LF in 36 out of 43 images. 3 LF were indiscernible or had atypical appearance, and 4 false detections were made on the lamina. Computation time was 4 seconds per image. The average error of the depth found by the automatic algorithm vs manual segmentation was 0.03 mm, with an RMS error of 0.63 mm, and Bland-Altman 95% limits of agreement of -1.2 mm to 1.3 mm; average error vs the sonographer was -2.7 mm, with an RMS error of 3.7 mm, and limits of -7.8 mm to 2.4 mm; and average error vs the needle insertion depth was 2.8 mm, with an RMS error of 5.4 mm and limits of -7.1 mm to 12.7 mm.
Discussion: The agreement between the automatic and manual measurements is comparable to the errors identified previously for ultrasound imaging . Most of the average error is estimated to arise from the inherent thick appearance of the LF on the ultrasound. The bone map may help emphasize the key structures in the ultrasound image for additional guidance, even if the depth is measured manually. In conclusion, the performance of this new algorithm suggests it could be useful as an aid for needle insertion, but correct image interpretation is still important.
1. Best Pract Res Clin Anaesthesiol 2005;19:175-200.
2. Anesth Analg 2007; 104:1188-92.
3. Anesth Analg 2009; 109: 661-667.
4. Ultrasound in Medicine and Biology 2009; 35: 1475-1487.