Articular cartilage defects: in vitro evaluation of accuracy and interobserver reliability for detection and grading with US.

Abstract

To determine the accuracy and reliability of detecting and grading articular cartilage defects in porcine and human knees by using ultrasonography (US). US was used to evaluate 175 porcine and 16 human knee surfaces with a linear 5-12-MHz transducer. Porcine defects of varying diameter and depth were surgically created. Each porcine surface was independently assessed in blinded fashion by two radiologists for the presence and severity of defects. Accuracy of detection, interobserver reliability, and concordance between US and surgical grades were determined. Human specimens were retrieved from knees of patients who underwent joint arthroplasty. Defects in human knees detected with US were correlated with defects seen at direct surface visualization. Sensitivities for detection of porcine defects were 94% and 93% for readers 1 and 2, respectively; specificities were 90% and 77%, respectively; positive predictive values were 98% and 95%, respectively; and negative predictive values were 78% and 73%, respectively. Interobserver agreement was high (weighted kappa = 0.80), and concordance between US and surgical grades for both readers was high (weighted kappa = 0.90 and 0.78). In human cartilage, the distribution of cartilage denudation determined at US was the same as that determined at direct visualization. High-frequency US was accurate and reliable for detection and grading of knee articular cartilage defects.