Search

This study focuses on the visual problem-solving process of clinical pathologists. Its aim is to find expertise-related differences in the temporal arrangement of this process, with a special focus on the orientation phase. A theoretical model of the visual diagnostic process of medical specialists is extended with general problem-solving theory. Participants were 13 experts, 12 intermediates, and 13 novices, who all diagnosed seven microscopic images. Their microscope movements and thinking aloud were recorded. To study temporal arrangement of the process, we applied a time-grid to the data. The results reflected several aspects of general problem-solving theory. Experts and intermediates showed a more extensive orientation phase and more refined schemata than novices. Intermediates also showed a control phase at the end of the diagnostic process. Novices showed a uniform process. These phases were reflected in microscope navigation and thinking aloud, which justifies the extension of the theoretical model.

Visual expertise is becoming more and more a crucial part of medical expertise. The amount of medical images produced has grown tremendously and the last decades have welcomed several new and increasingly dynamic imaging techniques (Krupinski, 2010). Nevertheless, visual expertise has so far mainly been studied in domains using static images, such as X-ray images (Reingold & Sheridan, 2011). This is also true for the domain of clinical pathology, where pathologists interactively navigate and zoom within digital slides. Research in this domain either restricted these navigation possibilities (Krupinski et al., 2006; Tiersma, Peters, Mooij, & Fleuren, 2003) or neglected the visual processes (Crowley, Naus, Stewart, & Friedman, 2003). Hence, not much is known on visual expertise in domains that deal with interactive, medical images.In this study, six clinical pathologists of different expertise levels, performed diagnostic tasks on a virtual microscope (i.e., computer software displaying scanned tissue samples in a similar way as a light microscope). By measuring eye movements, microscope navigation and thinking aloud, insight in both visual and cognitive expertise was gained. Each participant viewed four histological slides, representing different diseases of the colon and annotated diagnostically relevant areas in the slide, while thinking-aloud. At the time of writing this submission, the data collection was just finished, hence the analysis is not fully completed, yet. However, some expected results – based on theory and prior studies – are given in the extended summary. This study will contribute to the understanding of visual expertise by developing a methodology which is fit for the study of interactive images.

Visual expertise is becoming more and more a crucial part of medical expertise. The amount of medical images has grown tremendously and the last decades have welcomed several new and increasingly dynamic imaging techniques (Krupinski, 2010). Nevertheless, visual expertise has so far mainly been studied in domains using static images, such as X-ray images (Reingold & Sheridan, 2011). This is also true for the domain of clinical pathology, where, in their natural context, pathologists interactively navigate and zoom within microscopic slides. Research in this domain either restricted these navigation possibilities (Krupinski et al., 2006; Tiersma, Peters, Mooij, & Fleuren, 2003) or neglected the visual processes (Crowley, Naus, Stewart, & Friedman, 2003). Hence, not much is known on visual expertise in domains that deal with interactive, medical images. In this study, 38 clinical pathologists with three different expertise levels, performed diagnostic tasks on a virtual microscope (i.e., computer software displaying scanned tissue samples in a similar way as a light microscope). By measuring eye movements, microscope navigation and thinking aloud, insight in both visual and cognitive expertise was gained. Each participant viewed seven interactive slides, representing different diseases of the colon. At the time of writing this submission, the analysis was not fully completed, yet. However, both literature and a previous study in this project, using static images only, revealed that novices spent less time and fixations in diagnostically relevant areas. This study will contribute to the understanding of visual expertise by developing a methodology which is fit for the study of interactive images.

For the diagnostic process in clinical pathology, the first few seconds are of major importance. According to pathologists, this is the time period where an important part of the diagnosis is formed. To test this hypothesis, this experiment had pathologists of different expertise levels perform diagnoses with a time constraint (two seconds). 38 clinical pathologists of different expertise levels (13 pathologists, 12 residents, and 13 medical students; 14 males, 24 females) diagnosed 10 images of the colon. The images were stills from a digital microscope and thus two-dimensional colour pictures. Five of the images were low magnification images (approx. 3x), the other five high magnification images (20x). Two images regarded normal tissue, three adenomas, and five adenocarcinomas. Participants only saw the images for two seconds. Patient background information was given upfront, and afterwards participants were asked to diagnose the image in a standardised way (multiple choice with five options) and explain verbally. Dependent variables investigated so far are general performance (correctness of diagnosis), several basic eye movements measures, and eye movements in areas of interest. So far, performance and eye movement data have been analyzed. Results showed a significant expertise effect on performance (between novices and intermediates/experts, not between experts and intermediates), scanpath length, and the number of fixations in diagnostically relevant areas. Also, significant effects of the magnification level and the nature of the stimulus (i.e. kind of disease) were found on a number of variables, such as number of fixations, scanpath length, and average fixation duration, indicating a bottom-up effect of the stimulus. The verbal data will be used to refine the performance outcomes, and to investigate the differences in verbal descriptions of the visual patterns between novices, intermediates and experts.

Ondanks het cruciale belang ervan voor sommige beroepen, is er nog geen helder beeld van hoe visuele expertise zich ontwikkelt. Evenmin is er een goedonderbouwde didactiek voor het aanleren van zulke expertise. In dit voorstel voor een poster wordt een onderzoek beschreven dat zich richt op het beantwoorden van deze twee kwesties – expertiseontwikkeling en didactiek – en daarbij de klinische pathologie als casus neemt. Het is erop gericht de verschillen in zowel visuele als cognitieve expertise van klinisch pathologen van verschillende expertiseniveaus (beginner, gevorderde, expert) in kaart te brengen. Theoretisch kader wordt gevormd door een cognitieve theorie over medische expertiseontwikkeling. Het onderzoek is een pilot. De verzamelde data omvatten oogbewegingen en microscopienavigatiedata (beide voor visuele expertise) en hardop-denkprotocollen (voor cognitieve expertise).

In this study, we propose a computational diagnosis system for detecting the colorectal cancer from histopathological slices. The computational analysis was usually performed on patch level where only a small part of the slice is covered. However, slice-based classification is more realistic for histopathological diagnosis. The developed method combines both textural and structural features from patch images and proposes a two level classification scheme. In the first level, the patches in slices are classified into possible classes (adenomatous, inflamed, cancer and normal) and the distribution of the patches into these classes is considered as the information representing the slices. Then the slices are classified using a logistic linear classifier. In patch level, we obtain the correct classification accuracies of 94.36% and 96.34% for the cancer and normal classes, respectively. However, in slice level, the accuracies of the 79.17% and 92.68% are achieved for cancer and normal classes, respectively.