A Fully Automatic Measurement of Lumbar Multifidus Muscle Thickness from Ultrasound Image

Low back pain (LBP) is a common disorder involving the muscles and bones and about half of the people experience LBP at some point of their lives. Thus, the social economic cost and the recurrence rate are also high. Transversus abdominis (TrA) and multifidus muscle are most important muscles understanding and treating low back pain thus physiotherapists have increasing interest to them for clinical and research purposes. In this paper, we propose a new method for fully automated measurement of thickness for lumbar multifidus muscle from rehabilitative ultrasound image with a series of image processing techniques. The proposed method first extracts boundary lines of subcutaneous fat and boundary area between muscle and spine by applying Ends-in Search stretching, Grassfire algorithm, cubic spline interpolation, and 4-directional contour tracing. Then the representative spine is also extracted similarly but the measuring point is usually located in the "valley" of the spine area from the observation of human expert's decisions thus a valley search algorithm is applied to find the measuring point. The experiment using 100 convex DICOM ultrasound image with 532 x 434 size from 12 ordinary male subjects. The discrepancy from human expert in multifidus muscle thickness is very low-less than 0.3 cm in 78 of 100 cases and less than 0.2 cm in 57 out of 100 cases. While ultrasound imaging has many advantages over MRI/CT in cost-effectiveness and no radioactive risk, it has examiner inconsistency problem. By being fully automatic with minimal discrepancy from human expert, the proposed method can contribute to analyze important muscles without suffering examiner inconsistency in rehabilitative ultrasonography.