Establishing the level of digitization for wrist and hand radiographs for the third National Health and Nutrition Examination Survey

Abstract

In the third National Health and Nutrition Examination Survey (NHANES III) conducted by the National Center for Health Statistics, Centers for Disease Control and Prevention, radiographs of the hands and knees were taken of participants 60 years and older as part of the study of arthritis and musculoskeletal conditions. The purpose of the study was to decide the digitizing resolution to be used for these radiographs. A set of wrist and hand radiographs (N=49) was graded by two radiologists for degree of bone erosions and served as a “gold standard.” The radiographs were then digitized at three resolution levels; low-resolution 150μm (2001×1634×12 bit matrix); intermediate-resolution 100 μm (3000×2400×12 bit matrix); and high-resolution 50 μm (4900×3000×12 bit matrix). A comparison of the digital images versus the gold standard reading was made at the three resolutions by two radiologists. Kappa statistics suggested fair (K>.4) to excellent (K>.75) agreement between the gold standard and the images at all levels. Intraclass correlation coefficient suggested high agreement between readers (ICC>.5), with minimal individual reader effect. Variance component estimates showed that the major contribution (78–83%) to scoring came from variability in the images themselves, not from the readers. The 100μm resolution was selected over the 150 and 50 μm on the basis of practical considerations such as storage requirements, display time, and easier manipulation of the digital images by the readers.