S-Space College of Medicine/School of Medicine (의과대학/대학원) Dept. of Medicine (의학과) Theses (Ph.D. / Sc.D._의학과)
Non-intrusive Musculoskeletal Evaluation Using Inertial Sensor and Depth Camera
관성센서 및 깊이 카메라를 이용한 비간섭적 근골격계 평가
- 의과대학 의학과
- Issue Date
- 서울대학교 대학원
- motion analysis; inertial measurement unit; depth camera; musculoskeletal system; neck; shoulder; spine
- 학위논문 (박사)-- 서울대학교 대학원 : 의학과 의공학 전공, 2016. 8. 김희찬.
- New technologies, such as inertial measurement units (IMUs) and infrared depth camera (e.g. Kinect), have been introduced to detect and assess human motion. Whereas conventional motion analyses using optical markers and motion capture systems (MoCap) are intrusive, expensive, technically demanding, and restricted by time and place, the newer technologies are non-intrusive, easy to apply, and capable of real-time, continuous monitoring at moderate cost and effort. However, their clinical application is still limited because they have not been validated for clinical usage.
Motion analysis is pivotal for the evaluation of the musculoskeletal system because motion is the cause of a majority of musculoskeletal disorders and may also be used to treat several of these disorders. For these novel technologies to be applied clinically, their reliability, validity, and utility should be evaluated for ‘specific’ musculoskeletal conditions.
In this thesis, three individual studies were conducted. First, the validity of IMUs and Kinect for assessing cervical motion was tested in comparison with MoCap. Second, the clinical utility of Kinect for the diagnosis of adhesive capsulitis was evaluated. Thirdly, the practical application of IMUs for real-time detection of injurious movements during exercise was attempted.
In the first study, cervical motions were measured using Kinect, IMUs, and MoCap simultaneously and compared with each other. Fifteen healthy volunteers enrolled in the study. They performed three cardinal movements (axial rotation, flexion/extension, and lateral bending) with 2 IMUs attached on the vertex (head) and on the xiphoid (trunk), respectively
a Kinect was placed in front of the participants. The experiment was performed in a lab equipped with MoCap. The orientation of the head and trunk was calculated from each IMU or from outputs of the “software development kit (SDK) for face recognition” and “SDK for body posture recognition”, respectively. For the calculation of neck motion, a subject was modeled as 2 rigid bodies with a ball joint. Rotation matrices for neck motion were induced from orientation matrices for the head and trunk.
In the study evaluating the clinical utility of Kinect for the diagnosis of adhesive capsulitis of the shoulder, 15 healthy volunteers and 12 patients with adhesive capsulitis participated. Their active range of motion (ROM) and passive ROM of the shoulder was measured using a conventional goniometer for flexion, abduction, and external rotation of the shoulder. Thereafter, they performed the same active ROMs using a Kinect. The ROMs measured by Kinect was calculated using projection angles in the sagittal, coronal, and arm-axial planes. Their agreement with active/passive ROMs measured by a goniometer was assessed and their utility for the diagnosis of adhesive capsulitis was evaluated by Cohen’s kappa.
In the study assessing the real-time application of IMUs during exercise, 38 participants (18 with low back pain and 20 healthy volunteers) performed the squat with three IMUs attached on the lumbar and sacral spine and on the lateral thigh. “Spine hinge” movement (also known as “buttock wink” was detected by monitoring: 1) the abrupt increase of lumbosacral flexion based on the difference of rotations of the two IMUs in the lumbar and sacral spine (LS_hinge) or 2) change of direction of sacral rotation, from forward to backward, based on the IMU on the sacral spine (S_hinge). The rotation angle of the lumbar/sacral spine was calculated by simple subtraction of the orientations of each IMU around the mediolateral axis with time because the squat was presumed to occur exclusively in the sagittal plane. Clinical parameters (low back pain, range of passive hip flexion, and trunk muscle power) were assessed in the subjects.
In the first study, the 95% limit of agreement (LoA) between IMUs and MoCap showed a fair agreement (<10°) and LoA between Kinect and MoCap showed a less favorable agreement (>10°) for measuring ROM in all directions. Correlation between the MoCap and IMUs within the -40° to 40° range was excellent for flexion/extension and lateral bending [intraclass correlation coefficient (ICC) > 0.9], and fair for axial rotation (ICC > 0.8). Correlation between the MoCap and Kinect system within the -40° to 40° range was fair for all motions (ICC > 0.8).
In the second study, the ICC for flexion/abduction/external rotation between goniometric passive ROM and Kinect ROM was 0.906/0.942/0.911, while that between active ROMs and Kinect ROMs was 0.864/0.932/0.925. Cohen’s kappa values for the diagnosis of adhesive capsulitis were 0.88, 0.88, and 1.0 using the three cutoff values: the same as passive ROM values, reflecting the mean difference, and based on receiver operating characteristic curves.
In the third study, LS_hinge and S_hinge correlated highly [ICC = 0.84, Pearson Correlation Coefficient (PCC) = 0.85 with p <0.01]. Larger passive hip flexion range was related to later spine hinge occurrence (PCC = -0.361 and p = 0.03). In the wide-stance squat in healthy controls, spine hinge occurred at a deeper position than narrow-stance squat (p = 0.00, Wilcoxon signed-rank test). Greater trunk muscle power was related to lesser sacral rotation relative to lumbar rotation (PCC = 0.52, p = 0.00). Low back pain did not significantly affect spine hinge or mobilization pattern.
In summary, neck motion can be appropriately measured using either IMUs or the Kinect while axial rotation measurement was relatively imprecise than for other directions. Nevertheless, their LoAs are acceptable for the long-term monitoring of cervical posture. Shoulder ROMs measured by the Kinect excellently agree with those measured by a conventional goniometer, implying that the Kinect may be used to measure shoulder ROM for the diagnosis of adhesive capsulitis of the shoulder. A single IMU at the sacrum was equivalent to dual IMUs at the lumbar and sacral spine for detecting spine hinge. Spine hinge and motor pattern of the squat are associated with hamstring/iliotibial tract tightness and trunk muscle strength. Low back pain alone seems not to affect the occurrence of spine hinge during the squat.
To conclude, novel technologies such as IMUs or depth cameras involve different accuracies or validities for different applications. For the evaluation of the musculoskeletal system, they are acceptable for measurement, diagnosis, or real-time monitoring of human motions if they are appropriately employed and analyzed. Their applications should be validated for specific purposes and may thereafter, be used for practice.