Richard G. Swensson

Saccadic eye movements to peripherally discriminated visual targets.

Two experiments required subjects to identify a peripheral target embedded among nontarget stimuli and fixate it as quickly as possible with a single saccadic eye movement. Experiment 1 varied both the target distance and its angular position between trials; the mean oculomotor latency, the proportion of erroneous movements, and the proportion of (correct) movements followed by a corrective saccade all increased as a function of target distance. Experiment 2 held target distance constant (12.7 degrees) and used verbal instructions to manipulate the speed and accuracy of the subjects oculomotor performance between conditions. The speed/accuracy trade-off was similar for all subjects. The reduced uncertainty about target distance in Experiment 2 made each subjects oculomotor performance more efficient. Error trials not only included apparent perceptual errors (initial movements to nontarget stimuli) but also motor errors - that is, instances when the initial erroneous movement was followed, with an extremely short latency, by a large saccade to the target. The characteristics of these motor errors suggest that the saccade is not planned in terms of its amplitude and direction in retinal coordinates.

Lesion detection and signal-to-noise ratio in CT images.

This study measured observers’ ability to detect and locate focal lesions on simulated CTimages. The difficulty of the detection task was manipulated by changing the difference in attenuation between the lesion and its background (contrast), the random variation in the CT values (noise) or the lesion’s size. The human observers’ performance was compared to that of matched filter detector, modified to include the effects introduced by the display window and the uncertainty about the lesion’s location on the image. Changes in lesion contrast, lesion size and noise produced large variations in both the lesion signal–to–noise ratio (a measure of the matched filter detector’s performance) and estimated measures of the observer’s detection and localization ability. Changes in observers’ performance were closely related to changes in lesion signal–to–noise ratio. Generally, changes in lesion contrast, lesion size or noise that produced similar values of lesion signal–to–noise ratios had equivalent effects upon the observers’ performance.

Detection of noisy visual targets: models for the effects of spatial uncertainty and signal-to-noise ratio.

Brigham and Women’s Hospital, Boston, Massachusetts 02115 An “extreme-detector” model for detecting spatially uncertain targets in noisy backgrounds predicts how both detection and localization abilities are degraded by increasing the number of possible target locations. Experiments 1 and 2 show that the model accurately predicts detection and localization performance in tasks with two, four, and eight locations from d’ estimates of the observer’s ability to detect the target in a known spatial location. These predictions can be linked to the physical stimuli by combining the extreme-detector model with a “psychophysical” model that specifies how stimulus measures determine the target’s detectability in a given location. Single-parameter fits of four such combined models were compared with estimates of detection and localization performance in Experiment 3, which manipulated the target’s physical signal-to-noise ratio across various conditions of an eightlocation task.

A two-stage detection model applied to skilled visual search by radiologists

Harvard Medical School and Peter Bent Brigham Hospital, Boston, Massachusetts 02115 The model treats the detection of targets in a visual search task as a concatenation of two serial detection stages. Preattentive visual mechanisms in the initial stage function as a filter, selecting specific features of a visual pattern for the observer’s explicit attention and final cognitive evaluation. The model uses bivariate normal distributions to represent the decision variables for the two serial stages, assuming different parameters for the target and nontarget features in a test set. The model is applied to the detection performance of radiologists interpreting chest x-rays under various conditions of search. It accounts for the substantial improvement in radiologists’ ability to distinguish between target and nontarget test features when they had to search the x-ray images, compared to their performance without visual search. A change in the ROC curve between two different search tasks could be interpreted as a shift in the selection cutoff used by the preattentive filter.

Analysis of rating data from multiple-alternative tasks☆

Abstract A standard method of estimating a single ROC curve from rating data for two stimulus alternatives has been extended to ratings of multiple alternatives. An observers ratings are assumed to represent ordinal classifications of a unidimensional decision variable that has a separate distribution for each of M possible stimuli. From these rating data, a maximum-likelihood procedure simultaneously estimates the rating-category boundary values and the 2(M−1) distribution parameters that specify ROC curves between all pairs of the M stimulus alternatives. Many stimulus manipulations, particularly those investigated in psychophysical experiments with visual or auditory stimuli, could justify this M-alternative rating procedure and analysis. An advantage of this method is that it allows reliable measurement of an observers performance indices at much higher values than does the two-alternative method. The assumption of a unidimensional decision variable may be too restrictive for general decision-making situations, where the decisions among alternatives often involve multiple sources of information. However, the two-alternative method is commonly used to fit rating-ROC curves for some decision-making tasks, such as diagnosis from medical images, for which the multiple-alternative procedure actually might be more appropriate.

Improving Diagnostic Accuracy: A Comparison of Interactive and Delphi Consultations

Consultation among physicians on difficult diagnostic problems is commonly used to improve the accuracy of medical decisions. Such consultation is most often informal and interactive. Nevertheless, behavioral studies suggest that non-interactive techniques may be more effective problem solving methods. Of these the Delphi approach, involving pooling and feedback of anonymously contributed information, has generated particular interest. To assess the relative effectiveness of independent decision making, interactive group consultation, and Delphi techniques in a clinical setting we compared the diagnostic accuracy of 17 radiologists interpreting radiologic examinations in these settings. Interactive consultation improved performance by 69% compared to radiologists interpreting the studies individually. In addition, two Delphi strategies each produced an additional 20% mean improvement in accuracy over interactive consultation. Whereas interactive consultation improved the accuracy of the best individual readers by only 6%, a Delphi model improved their performance by 25%. Thus, Delphi was an effective, easily applied method of clinical consultation whose usefulness in other clinical setting should be evaluated.

Detection of small focal lesions in CT images: effects of reconstruction filters and visual display windows

The detectability of small, high-contrast lesions was measured on CT images, simulations of those obtained by the EMI Mark I scanner. Images were reconstructed using five reconstruction filters (kernels), which varied the image sharpness and noise level. Different sets of images were produced using various CT display windows, six different window sizes and four different display level settings. The measured lesion detectability for observers increased from 1.6 to 2.4 as the reconstruction kernel became smoother, and it decreased only slightly at the largest display window (1000 CT numbers wide). These effects were predicted by changes in the signal-to-noise ratio, as calculated for the lesion-matched filter applied to each set of physical CT images. This filter computes the cross-correlation of the CT image and the lesion profile at the specified possible locations for the lesion.

Observer variation and the performance accuracy gained by averaging ratings of abnormality.

Six radiologists used continuous scales to rate 529 chest-film cases for likelihood of five different types of abnormalities (interstitial disease, nodule, pneumothorax, alveolar infiltrate, and rib fracture) in each of six replicated readings, yielding 36 separate ratings of each case for the five abnormalities. Separate data analyses of all cases and subsets of the difficult/subtle cases for each abnormality estimated the relative gains in accuracy (linear-scaled area below the ROC curve) obtained by averaging the case-ratings across (a) six independent replications by each reader (25% gain), (b) six different readers within each replication (34% gain), or (c) all 36 readings (48% gain). Although accuracy differed among both readers and abnormalities, ROC curves for the median ratings showed similar relative gains in accuracy, somewhat greater than those predicted from the measured rating correlations. A model for variance components in the observers latent decision variable could predict these gains from measured correlations in the single ratings of cases. Depending on whether the models estimates were based on realized accuracy gains or on rating correlations, about 48% or 39% of each readers total decision variance (summed variance for positive and negative cases) consisted of random (within-reader) error that was uncorrelated between replications, another 10% or 14% came from idiosyncratic responses to individual cases, and about 43% or 47% was systematic variation that all readers found in the sampled cases.

On the relations between random walk models for two-choice response times ☆

Abstract We state a condition on the moment-generating functions of the random walk model derived by Link and Heath (1975) , which is necessary and sufficient to yield the sequential probability ratio test of Wald (1947) and the reaction-time model proposed by Stone (1960) . A restricted form of this condition can also be obtained by imposing the types of symmetries on the moment-generating functions considered by Link and Health (1975) and Thomas (1975) . The various conditions are contrasted and their implications for empirical investigation are discussed.

Nodule polarity effects on detection and localization performance in liver CT images

Performance accuracy for detecting and localizing small nodules on liver CT images depends on whether an observer is required to find dark nodules or bright nodules on those images. We investigated these asymmetric polarity effects using simulated nodules of varying sizes placed on spiral CT scans of clinical patients acquired with intravenous contrast material, which made blood vessels appear brighter than liver background on the displayed CT images. A concurrent analysis of each observers detection-rating and scored-localization data estimated separate perceptual effects for the nodules of different sizes, and for locations of the dark or bright hepatic findings that observers regarded as most suspicious on the CT images. The results were consistent with equal visibility for dark and bright nodules of identical size and CT-contrast, and a linear increase in visibility with nodule signal-to-noise ratio for a non-prewhitening matched-filter calculation (NPW-SNR). The substantial lower accuracy for detecting and localizing the bright nodules, compared to the dark nodules, was a polarity effect apparently produced by the non- stationary liver CT backgrounds -- i.e., the presence of stronger confusing signals from the bright hepatic findings on these (contrast-enhanced) CT images than from the dark hepatic findings.