INTRODUCTION

Computer assisted motion analysis systems provide an objective tool for recording and evaluating human movement. It is important that the validity of motion analysis systems be established in order to consider the data of value to clinicians. Various sources of error exist with camera-based systems, but the contributions of each depend upon which system is being used. Previous research has quantified single sources of error; however, clinicians are concerned with the total errorin the calculated joint angle. This study was conducted to examine the total error associated with joint angle calculations using computer assisted motion analysis.

The objective of this study is to determine the criterion reference validity of the APASTM and GaitLabTM software systems for the calculation of lower extremity joint angles.

REVIEW AND THEORY

Figure 1: Flowchart for Joint Angle Calculation

Figure 1 is a flow chart of the steps for camera-based analysis of joint angles. At each level of this flowchart, error may be introduced into the process. Errors are introduced due to marker misplacement or soft tissue movement over bony prominences2,5,6,8,11. Soft tissue error in the lower extremities of 10-20 mm2, and soft tissue error in the ankle of 8.8-12.8 mm11 have been reported. Subject motion is recorded using cameras, and the two-dimensional camera recordings of marker locations are transformed into three-dimensional coordinates of the markers in the laboratory space. Errors introduced at this point may be due to associated hardware (cameras, lenses, calibration frame), or due to the direct linear transformation7,12,15. RMS error of 5.7 mm in a calibration field 3.5 meters long, 2.5 meters high, and 1.5 meters wide has been reported with direct linear transformation15.

The coordinates of the markers may be analyzed using a simple point-to-point method, or using a model. The simple analysis connects marker locations using vectors, and then calculates angles between these vectors (Fig. 1, left branch of flowchart). Calculation errors related to the inverse trigonometric functions increase as the angle approaches 180?,13. A major limitation of this simple analysis is that motions out of the plane of the markers cannot be measured. In order to measure out of plane motion, a model must be used.

A more complex model incorporates a segmental coordinate system and anthropometric measurements to estimate joint centers, or a body-centered coordinate system can be imbedded at a body segment's center of mass to calculate three dimensional joint angles. Errors may arise from inaccurate anthropometric measurements, or when extrapolating beyond the limits of the data from which the model was created. In addition, errors may occur with joint center estimation, especially at the hip1,3,4,6,9. When the joint angle approaches 180? the cosine algorithm again introduces error13.

While these researchers have looked at each of the individual errors, clinicians are more interested in the total error. Stanhope developed a test procedure that was effective in quantifying the magnitudes of angular kinematic inaccuracies associated with unilateral lower extremity gait analysis12. A series of tests were performed on an idealized lower extremity model, to estimate errors in calculated knee and ankle angles. This test allowed for evaluation of a composite error from sources including hardware, direct linear transformation, segmental coordinate systems, and estimation of joint centers. Kinematic inaccuracies were quantified as the mean and range of knee and ankle angular displacement. Joint displacement error including sources such as camera obstruction and model motion were found to be 3.6? However, this composite error did not account for errors due to soft tissue movement, marker misplacement, joint motion in the multiple planes, and motion at multiple joints. In addition, the lower extremity model used did not allow for articulation at the knee, nor did it include the hip joint.

In this study, the left lower extremity of a plastic skeleton was fixed in various positions using clamps and rods. Each position combined hip and knee flexion and extension, and hip ab/adduction. Kinematic data were collected and joint angles were calculated using two methods: the software provided with the Ariel Performance Analysis SystemTM and GaitLabTM gait analysis software14. These were compared to direct angular measurements taken with an electronic inclinometer to evaluate total composite error.

PROCEDURES

Data were collected using the Ariel Performance Analysis SystemTM, a video based three-dimensional motion analysis system. A calibration device consisting of six steel cables and twenty-four reflextive markers was used to define a control volume 1.2 m wide x 2.0 m long x 1.7 m high. Nine retroreflective markers were placed on the pelvis and left lower extremity of a plastic skeleton. The skeleton was fixed, in approximately 10?increments, from 13?of hip extension to 40?of hip flexion and in approximately 5?increments from 22?of abduction to 11?of adduction. The knee joint angle was fixed from 4?to 83?of flexion in approximately 10?increments. After the skeleton was fixed in a position, the joint angles were measured by three researchers using an electronic inclinometer. This process was repeated three times to yield a total of nine measurements per joint in order to establish intra- and interrater reliability. The skeleton was videotaped while moving through the calibrated field, then the joint angles were remeasured by one researcher to assure that there was no change in position.

The images from each videotape were grabbed and, automatically or manually digitized using the APASTM. Manual digitization was performed only when the system misidentified markers. The direct linear transformation was used to calculate 3-D marker locations. The marker data was downloaded from APASTM, then analyzed using the APASTM software and GaitLabTM to calculate the joint angles.

RESULTS

Intraclass correlation coefficients (3,1) were calculated for each joint motion to establish intrarater reliability for each researcher, and to establish interrater reliability of measurements obtained with the electronic inclinometer for each joint motion. All values were greater than .995.

Errors were calculated for each method throughout the range of angles (Table 1).

Error range	GaitLab	APAS
Hip Flexion/Extension	0.1 - 18.6	0.6 - 17.8
Hip Ab/Adduction	0.2 - 7.2	32.0 - 97.5
Knee Flexion/Extension	1.2 - 11.9	1.3 - 6.1

Table 1: Ranges of error for specific joint angles

DISCUSSION

For most trials, the APASTM and GaitLabTM hip and knee flexion and extension angle measurements both agree with the inclinometer; however, occasionally, there are large, non-systematic disagreements of up to 18.6? between the inclinometer, as the standard, and the video analysis software interpretations of joint angles. This occurred near neutral hip flexion/ extension, as measured by the inclinometer. As only two data points were collected in this position, conclusions cannot be drawn.

When the error for each plane of motion for both software programs was plotted against the range of motions that were tested, no systematic error among the joint positions tested could be ascertained. The only consistent error among the calculated joint angles occurred with hip abduction and adduction using the APASTM software. These errors range from 32.0?to 97.4? This was not unexpected, since APASTM calculates angles using marker to marker vectors, and no marker can be placed on the actual hip center.

In this study, we found a larger errors than has been published in previous studies. These studies, not based on an anthropometric structure, have shown video based analysis systems to be accurate to within a few degrees. An under estimation of 2.4?was reported at 180?by Vander Linden et al.13 Mean angular error of 0.26?was reported by Klein7 , and error of 2.1?was reported by Scholz10. All of these studies used retroreflective markers attached to goniometers under static conditions. This does not simulate the human structure to a large degree, and lacks applicability to clinical situations. Our study depicts the accuracy of these systems as they are routinely used, and has more clinical relevance and validity. Future studies should continue to address total error, possibly incorporating an estimate for soft tissue movement.

REFERENCES

1. Campbell KR, et al. J Biom, 21:860. 1988.

2. Cappozzo A., et al. Proc. ISB XIVth Cong, 238-9, 1993.

3. Duke RP, et al. J Biom,; 10:781-786, 1977.

4. Grood ES, et al. J Biomech Eng.,105:136-144, 1983.

5. Kadaba MP, et al. J Orthop Res. 7(6):849-860, 1989.

6. Kadaba MP, et al. J Orthop Res. 8:383-392, 1990.

7. Klien PJ, et al. Arch Phys Med Rehab. 76:183-189, 1995.

8. O'Connor PD, et al. Phys Ther. 73(7): 478-483, 1993.

9. Ramakrishnan HK, et al. J Biomech, 10: 969-977, 1991.

10. Sholtz JP. Phys Ther, 69(8): 679-689, 1989.

11. Stanhope SJ. Gait & Posture, 2 (1): 58, 1994.

12. Stanhope SJ. Proc. 13th Southern Biomed. Eng. Conf., Wash. D.C., 1030-1034, 1994.

13. Vander Linden DW, et al. Phys Ther., 72(4):300-305, 1992.

14. Vaughan CL, et al. Gait Analysis Laboratory. Human Kinetics Publishers, 1992.

15. Wood GA, et al. J Biom. 19:781-785, 1988.

ACKNOWLEDGMENTS

We would like to acknowledge Dr. Stephen Stanhope, Ms. Christine DelMarcelle, Dr. Theodore Bross, Mr. Michael Hopwood and Mr. Peter Guzzetti for their assistance with this research.