OBJECTIVE
The objective of our study was to compare child abuse detection using screen-film radiographs and their digitized images displayed on a computer workstation.
MATERIALS AND METHODS
Skeletal surveys of 20 consecutive child abuse patients whose abuse was clinically proven by a combination of history, physical and radiographic findings, and social work history, and 20 consecutive control subjects were evaluated. Three radiologists rated both the screen-film radiographs (400-speed, double-emulsion film) and their digitized images displayed on a workstation (2K x 2K resolution) using a six-point ordinal scale for suspicion of child abuse, fracture detection, and image quality. The rating response was analyzed using multiobserver-multicase receiver operating characteristic analysis of variance. The McNemar test was used to evaluate differences between imaging techniques and between diagnoses made using each imaging technique and clinically proven child abuse.
RESULTS
The area under the receiver operating characteristic curve for screen-film radiographs was 0.934+/-0.025 and for digitized images was 0.922+/-0.013. This difference was not significant (p = .658); however, two observers significantly underestimated the child abuse diagnosis with digitized images (p = .02). In a review of the false-negative child abuse diagnoses, observers failed to recognize characteristic metaphyseal fractures (10 observations) and rib fractures (five observations) on digitized images that had been recognized on screen-film radiographs. Mean image quality was rated significantly lower (p < .0001) and interpretation time was significantly longer (75 sec; p < .001) for the digitized images than for screen-film radiographs.
CONCLUSION
The characteristic types of fractures that were not identified on the digitized images, lower image quality, and longer interpretation time raise concern that digitized images may not be adequate for interpretation of suspected child abuse.