THE EFFECT OF AUTOMATION ON THE REPEATABILITY OF MEASURING ACETABULAR WEAR

Abstract

The present study investigates the repeatability of two new methods of measuring acetabular wear with differing levels of automation. Experimental evaluation showed that the more automated method was more repeatable. Both methods segmented the femoral head and acetabular rim with ellipses. The displacement of the ellipse centres was measured and the difference at year 1 and 5 taken as a measure of wear. Measurements were obtained twice for each case. The less automated of the two methods involved the annotation of 9 points on the femoral head and 18 on the acetabular rim to which two least squares ellipse fits[1] were performed. The second and more automated method was active ellipses[2][3]. This method uses iterative robust ellipse fitting and a model of appearance learned from a training set to cause two ellipses to converge on the contours of the femoral head and acetabular rim from a single starting point. Fifty cases with radiographs taken at year 1 and year 5 were measured by both methods. The radiographs contained CPTs with 28mm heads and were digitized at 150 dpi. Fifty postoperative radiographs containing 22.225mm Zimmer CPT heads trained the more automated method. None of the radiographs had metal backed cups or highly eccentric rims. The repeatability coefficient (2 standard deviations) of the active ellipses was 0.23mm and that of the best annotator was 0.40mm while the worst was 2.69mm due to an outlying measurement. Limits of agreement were calculated between the two methods as −0.61mm to 0.91mm and show the active ellipses could replace annotation. Given that the active ellipses are nearly twice as repeatable this is desirable. The range of difference in measurements for the active ellipses is less than that of the annotator.