Author: William R. Oliver, M.D.
This report presents the results of the original methodology for evaluating the effect of image processing and image quality on the ability of forensic pathologists to accurately interpret images of patterned injury of the skin, and then describes revisions and results of the methodology due to findings from the initial study design.
The initial study design consisted of three surveys: the first to be a collection of "classic" images that most pathologists would diagnose with high consensus (baseline survey); the second to consist of degraded images with lesser resolution poorer composition, etc., to determine how degradation affected diagnostic consensus; and the third presenting images treated with various enhancement techniques (primarily contrast improvements) to determine whether any benefit was gained. Surprisingly, the first survey produced a median of only 74 percent consensus.
This led to a modification of the remaining surveys to determine the reason for the unexpectedly low consensus for the first survey. The second survey was modified to query respondents about why they did not reach what was thought to be an obvious consensus; and the third survey tested the effect of providing history and context for the observed injuries.
The second survey indicated that the primary reason participants did not join the consensus interpretation was due to perceived ambiguity due to the lack of a history.
An analysis of the third survey demonstrated the importance of context and history in forensic pathologic diagnosis. When provided with history, consensus rose to approximately 98 percent per question (median value) for the matching subset of the first survey.