SATISFACTION OF SEARCH IN DIAGNOSTIC RADIOLOGY

点击翻译按钮获取中文摘要

Underreading error (false-negative interpretation) is the most frequent and costly type of error occurring in diagnostic radiology. A subset of underreading error has been attributed to "satisfaction of search" which occurs when a second lesion goes unreported due to the detection of an initial lesion in a radiograph (Tuddenham, 1962,1963; Smith, 1967). Previous research supported by two NIH awards studied satisfaction of search (SOS) in the experimental laboratory to gain some understanding of this error in the hope of reducing its frequency. These experiments showed that satisfaction of search could be demonstrated in a rigorous way using ROC methodology with three types of radiographic examinations and abnormalities. In the first experimental procedure, detection accuracy for various subtle lesions in chest radiographs was compared with the accuracy for detecting the same lesions when a pulmonary nodule was added to the radiograph by simulation. Detection was reduced with the addition of artificial nodules. Two other experimental procedures were developed to represent a wider range of radiographic examinations in which clinical examples of SOS are common. In the second procedure, multi-trauma patients were depicted in a series of radiographs. Detection of subtle fractures was substantially reduced when other fractures were included in other images of each series. The third procedure mimicked a condition known to be error prone, on contrast studies of the abdomen where the contrast material itself may serve as a distraction. Reports of extra-intestinal abnormalities visible without contrast in abdominal radiographs were less frequent in contrast examinations as a consequence of significant increases in decision criteria. Unfortunately, the causes of SOS remain largely unknown. However, these preliminary studies provide a framework in which definitive research can address the complex questions that remain. Cases from the ROC studies that exhibited strong SOS will be studied using eye position recording. The processes underlying SOS may be reflected in two important measures of search, gaze dwell time on lesions and time for gaze to encounter lesions, that are available from eye position data. Three new experiments (one for each SOS procedure) will use gaze dwell time to determine whether satisfaction of search effects are due to errors in visual scanning, recognition, or decision making using a classification scheme developed by Nodine & Kundel (1987). A causal model of SOS that relates gaze encounter time and gaze dwell time to detection will be tested. By determining the causes of SOS, the nature of interventions likely to reduce SOS error and effective approaches to computer-aided diagnosis can be identified.