Impact of clinical history on fracture detection with radiography

The effect of knowledge of localizing symptoms and signs in the detection of fractures was studied. Forty radiographs of the extremities were examined twice by seven radiologists; the sessions were separated by 4 months. In 26 cases, a subtle fracture was present; 14 cases were normal. In half of the cases at each session, the precise location of pain, tenderness, or swelling was provided. The observer was asked to determine if the case was normal or abnormal (provide the exact location of the fracture) and to indicate the degree of confidence in the diagnosis. Responses were converted to a numeric scale for analysis. Analysis of receiver operator characteristic parameters indicates that clues regarding location of trauma facilitate detection of fractures. The improvement is based largely on an increased true-positive rate without an increased false-positive rate, regardless of the decision criteria of the radiologist (overall willingness to "overread" or "underread"). This has direct clinical applicability and reinforces the plea of radiologists for precise clinical information.     

Time course of satisfaction of search

"Satisfaction of search" (SOS) refers to the effect in which a second lesion remains undetected after detection of another lesion on the same radiograph. The objective of this study was to clarify our understanding of SOS by relating it to total time of inspection and time intervals before, between, and after discovery of lesions. Detection accuracy of native lesions in chest radiographs, before and after the addition of a simulated nodular lesion, was measured for ten observers. Analysis of data from this and a previous experiment showed that average perceptual accuracy of individual receiver operating characteristic curves was significantly reduced with the addition of the nodules. Plots and analyses of search time revealed that, on average, during a typical 46-second inspection of a case, simulated nodules were found at 18 seconds, native abnormalities at 25 seconds, and false positives occurred at 33 seconds. Time needed to find nodules did not depend on whether native lesions were present; time to find native lesions did not change with addition of nodules; and total search time was the same for images with one, two, or no lesions. The detection results show that the SOS effect was obtained, but that interrupting search in order to measure it also diminishes accuracy. Analysis of the time course data relates SOS to perceptual capture and strategic halting of search.     

Contribution of individual projections alone and in combination for radiographic detection of ankle fractures

OBJECTIVE: We wanted to determine whether the standard three-view ankle radiographic series could be replaced by a two-view combination, and if so, which two-view combination (anteroposterior with lateral or mortise with lateral) would be superior. MATERIALS AND METHODS: During a 12-month period, we retrospectively reviewed 556 consecutive ankle radiographic studies consisting of anteroposterior, mortise, and lateral views. One hundred twenty patients with at least one ankle fracture were paired with 140 healthy control subjects. Each image in the three-view examination was separated and sorted by view and studied independently; all images were reviewed by two skeletal radiologists and two orthopedic surgeons. Each radiograph was evaluated for fracture of the medial, lateral, and posterior malleoli and the foot using a five-point confidence rating. Performance of each view and modeled two- and three-view combinations of views was evaluated with modified receiver operating characteristic analysis. RESULTS: The data provide little support for preferring either two-view combination (anteroposterior-lateral or mortise-lateral) for any type of fracture. The three-view combination does detect significantly more fractures than some two-view combinations in some locations, and there is a statistically significant cost in diagnostic accuracy for eliminating the anteroposterior or mortise view. CONCLUSION: Reducing the ankle radiographic series from three to two views would result in a small but significant decrease in the detection of fractures of the ankle and foot. Both two-view combinations are equivalent for fracture detection.     

The utility of 99mTc depreotide compared with F-18 fluorodeoxyglucose positron emission tomography and surgical staging in patients with suspected non-small cell lung cancer

STUDY OBJECTIVES: The findings from conventional imaging modalities, such as chest CT, are frequently unreliable in patients with lung cancer. This study was designed to compare the relative diagnostic accuracies and utility of the two most widely used functional imaging examinations, F-18-2-fluoro-2-deoxyglucose (FDG) positron emission tomography (PET) and (99m)Tc depreotide scintigraphy, for the diagnosis and staging of lung cancer. DESIGN: Prospective, experimental investigation. SETTING: Academic medical center. PATIENTS: One hundred sixty-six subjects with suspected lung cancer were enrolled in the study. INTERVENTIONS: Whole-body and single-photon emission CT imaging of the chest was performed after IV administration of (99m)Tc depreotide. Attenuation-corrected FDG PET imaging was performed after IV administration of FDG. Image findings were compared with the biopsy results or clinical follow-up. Measurements and results: In 157 subjects with evaluable lung lesions, the sensitivities and specificities for detecting malignant disease (95% confidence intervals) of FDG PET are 96% (90 to 98%) and 71% (54 to 85%), and of (99m)Tc depreotide are 94% (88 to 98%) and 51% (34 to 68%). In the 139 subjects with available complete staging data, FDG PET correctly staged 76 of 139 patients (55%), and (99m)Tc depreotide correctly staged 63 of 139 patients (45%). CONCLUSIONS: The sensitivity for detection of lung cancer in the primary lesion is equally high for FDG PET and (99m)Tc depreotide. The specificity is superior for FDG PET. The staging accuracy of FDG PET and (99m)Tc depreotide is similar, but when read with the chest CT neither scintigraphic examination is sufficiently accurate to stage patients with non-small cell lung cancer.     

Perception testing in surgical pathology

A test predictive of mature pathologic expertise could be of great value in the selection of individuals entering the field. Some individuals may have an aptitude superior to others to discriminate images based on the discovery of distinctive image features. This may enhance their ability to draw pathologic conclusions from clinical images and may favorably affect their performance as pathologists. A form of the Facial Recognition Test was administered to pathology residents and faculty to test this hypothesis. Facial recognition results correlated well with resident performance measured by overall faculty ratings, but only marginally with performance measured by scores on "direct recognition" slides of the American Society of Clinical Pathologists' in-service pathology exams. These preliminary results indicate that testing of visual discrimination may be predictive of resident performance in pathology. A test of this type could eventually be useful in selection and self selection of resident candidates and in development of remedial training for residents experiencing difficulty in learning morphologic interpretation.     

Improving the accuracy of mammography: volume and outcome relationships

BACKGROUND: Countries with centralized, high-volume mammography screening programs, such as the U.K. and Sweden, emphasize high specificity (low percentage of false positives) and high sensitivity (high percentage of true positives). By contrast, the United States does not have centralized, high-volume screening programs, emphasizes high sensitivity, and has lower average specificity. We investigated whether high sensitivity can be achieved in the context of high specificity and whether the number of mammograms read per radiologist (reader volume) drives both sensitivity and specificity. METHODS: The U.K.'s National Health Service Breast Screening Programme uses the PERFORMS 2 test as a teaching and assessment tool for radiologists. The same 60-film PERFORMS 2 test was given to 194 high-volume U.K. radiologists and to 60 U.S. radiologists, who were assigned to low-, medium-, or high-volume groups on the basis of the number of mammograms read per month. The standard binormal receiver-operating characteristic (ROC) model was fitted to the data of individual readers. Detection accuracy was measured by the sensitivity at specificity = 0.90, and differences among sensitivities were determined by analysis of variance. RESULTS: The average sensitivity at specificity = 0.90 was 0.785 for U.K. radiologists, 0.756 for high-volume U.S. radiologists, 0.702 for medium-volume U.S. radiologists, and 0.648 for low-volume U.S. radiologists. At this specificity, low-volume U.S. radiologists had statistically significantly lower sensitivity than either high-volume U.S. radiologists or U.K. radiologists, and medium-volume U.S. radiologists had statistically significantly lower sensitivity than U.K. radiologists (P<.001, for all comparisons). CONCLUSIONS: Reader volume is an important determinant of mammogram sensitivity and specificity. High sensitivity (high cancer detection rate) can be achieved with high specificity (low false-positive rate) in high-volume centers. This study suggests that there is great potential for optimizing mammography screening.     

Contrast reduction by small localized stimuli: Extensive spatial spread of above-threshold orientation-selective masking

Striped gratings and a bullseye (concentric circles) were masked by stimuli that look unlike those patterns, but share orientational and spatial-frequency components with them. A small black dot reduced the apparent contrast of the entire bullseye; a thin black bar did so only to gratings with the same orientation. A bar reduced the apparent contrast of a small patch of grating across a retinal separation of over 4°. The orientation selectivity of the masking implicates mechanisms sensitive to patterned stimuli. The spread over a large retinal area suggests that some aspects of visual processing may be distributed (for example, Fourier-analytic) rather than local (space-domain).     

Multireader, multicase receiver operating characteristic analysis: an empirical comparison of five methods

RATIONALE AND OBJECTIVES: Several statistical methods have been developed for analyzing multireader, multicase (MRMC) receiver operating characteristic (ROC) studies. The objective of this article is to increase awareness of these methods and determine if their results are concordant for published datasets. MATERIALS AND METHODS: Data from three previously published studies were reanalyzed using five MRMC methods. For each method the 95% confidence intervals (CIs) for the mean of the readers' ROC areas for each diagnostic test, the P value for the comparison of the diagnostic tests' mean accuracies, and the 95% CIs for the mean difference in ROC areas of the diagnostic tests were reported. RESULTS: Important differences in P values and CIs were seen when using parametric versus nonparametric estimates of accuracy, and there were the expected differences for random-reader versus fixed-reader models. Controlling for these differences, the Dorfman-Berbaum-Metz (DBM), Obuchowski-Rockette, Beiden-Wagner-Campbell, and Song's multivariate Wilcoxon-Mann-Whitney (WMW) methods gave almost identical results for the fixed-reader model. For the random-reader model, the DBM, Obuchowski-Rockette, and Beiden-Wagner-Campbell methods yielded approximately the same inferences, but the CIs for the Beiden-Wagner-Campbell method tend to be broader. Ishwaran's hierarchical ROC sometimes yielded significance not found with other methods. Song's modification of DBM's jack-knifing algorithm sometimes led to different conclusions than the original DBM algorithm. CONCLUSION: In choosing and applying MRMC methods, it is important to recognize: (1) the distinction between random-reader and fixed-reader models, the uncertainties accounted for by each, and thus the level of generalizeability expected from each; (2) assumptions made by the various MRMC methods; and (3) limitations of a five- or six-reader study when the reader variability is great.