01whole.pdf (8.71 MB)
Expertise in visual search of medical and non-medical images
thesisposted on 2022-03-28, 13:04 authored by Ann J. Carrigan
This thesis aims to understand the crucial cognitive mechanisms that underpin visual search in medical images and the influence of expertise. Research suggests that experienced radiologists use information from an initial glance at an image to set the basis of their diagnosis. This research explores the information that can be extracted from images regarding the presence and location of a target. I use two stimulus types: real medical images on which I test both novices and expert radiologists, and natural scenes as a model for radiologist search. In Chapter 1, I present an overview of the literature relevant to these aims. Chapter 2 presents two experiments where I showed that a target could be both detected and located in a natural scene after a brief presentation (33 ms), but that visual clutter interferes with performance in a predictable way. In Chapter 3, I showed radiologists were able to detect and localise an abnormality in a mammogram presented for 250 ms at levels better than guessing.Crucially I demonstrated that a normal patient variant, high levels of breast density,affects performance. I conducted an in-depth analysis which emphasises the importance of considering factors such as stimulus variability, response imprecision, and participant guessing. In Chapter 4, I investigated the extent to which expertise guides attention based on prior experience with the prevalence of cancer, using a novel cueing paradigm where a chest radiograph (with or without a suspicious nodule) formed a prime. For naïve observers, an artificially boosted nodule in the prime radiograph guided attention, validating the task. Radiologists viewing true, more subtle nodules did not show the same effect, nor did they show any attentional guidance from cancer prevalence. However, more experienced radiologists seemed to be more sensitive to the subtle nodules than less experienced radiologists, suggesting that viexpertise might boost nodule salience. Finally, in Chapter 5, the implications of these findings are discussed in a broader context along with suggestions for areas of future research. Overall, my research shows that there is a large amount of information available after observers first look at a scene or medical image; more than previously thought. Further, the visual complexity of the display affects performance. Together, the experiments presented in this thesis advance the scientific understanding of the type of information available in the first glance and has clear implications for radiologist teaching and clinical benefits.