The Effects of Cue Precision on Medical Imaging Interpretation - PROGRAM SUMMARY: It is undesirable for radiologists to search for only one specific thing at a time, as they may miss unexpected findings. Despite this, there is evidence that prior knowledge and expectations (from clinical history or other cues) alter radiologists’ search performance and diagnostic accuracy. As Artificial Intelligence tools are added to the radiologic workflow, understanding the effects of different cues on radiologists will be critical for optimizing the way information is presented to radiologists. Cues that precisely describe targets (“find the red apple”) are known to guide search more efficiently than imprecise cues (“find the apple”). However, this precision effect has only been explored in cases where hue differentiates targets from distractor objects, which is not the case in radiology. And location cues (indicating where the target is) might increase both correct responses and false alarms, though it is unclear what precisely drives those false alarms. Filling these gaps in knowledge is critical for optimizing current and future systems that provide information to radiologists and other medical imaging professionals. In a novel approach, this project will quantify the effects of precise vs imprecise text cues and precise vs imprecise cues to target locations in radiologic search tasks while recording radiologists’ eye movements. The central hypothesis is that expert radiologists will search more effectively for abnormalities and will identify nodules more quickly when provided with more information (precise features and/or location) cues. The proposed project will precisely characterize the performance and gaze behavior of expert radiologists as they view chest x-rays ( ) and 3D volumetric CTs ( Aim 1 Aim 2 ). The research team will assess the effects of cue precision and cues indicating nodule locations on accuracy in identifying abnormalities, time taken to look at abnormalities, and changes in the radiologists’ overall strategies and other factors relevant to diagnostic accuracy. The results promise to transform current thinking about medical image viewing, where current models typically assume that targets are either completely unknown or are known with exacting precision. The results will also clarify and resolve long-standing mixed findings in the radiologic search literature: various studies have found that prior knowledge (e.g., clinical history or cues indicating what abnormalities to search for) either improves, decreases, or has no effect on diagnostic accuracy; might or might not influence later parts of the search task (such as time spent identifying an abnormality after looking at it); and either does or does not increase false alarms. By assessing eye movement dynamics, the proposed project will clarify the underlying mechanisms driving these effects. In addition, this project will provide opportunities for undergraduate students to engage in biomedical research and will greatly strengthen the biomedical research environment at NYIT, which has received limited NIH support to date.
