To what extent does nonanalytic reasoning contribute to visual learning in cytopathology?

Abstract

BACKGROUND Acquisition of visual interpretation skills in cytopathology may involve 2 strategies. Analytic strategies require trainees to base their interpretive decisions on carefully considered and often exhaustive cytomorphologic feature lists, a process that can be time‐consuming and inefficient. In contrast, nonanalytic pattern recognition strategies are rarely encouraged during training, even though this approach is characteristic of expert diagnostic behavior. This study evaluated the potential role of nonanalytic learning in cytopathology as an efficient alternative to analytic training. METHODS Forty‐nine cytology novice participants undertook an initial image interpretation test to obtain baseline diagnostic accuracy (test 1). Twenty‐four participants subsequently received training in basic cervical cytomorphology and were given a list of cell features for future reference (the “analytic” group). The remaining 25 participants were simply shown 20 nonannotated paired images of normal and abnormal cervical cells (the “nonanalytic” group). Following a practice phase, both groups were retested (test 2). Prior to a final test (test 3), participants in both groups were instructed to adopt a combined analytic/nonanalytic diagnostic strategy. Diagnostic accuracy and response times were measured in each test. RESULTS Diagnostic accuracy in both groups improved significantly between tests 1 and 2 (P<.001) but decreased between tests 2 and 3 (P<.05). Speed of response to test images was generally faster under nonanalytic than under analytic conditions. CONCLUSIONS Nonanalytic reasoning in cytopathology image interpretation can be as accurate as traditional feature‐based reasoning. Encouraging trainees to adopt pattern recognition strategies may help to expedite the acquisition of visual interpretation skills in cytopathology training programs, yet combining analytic and nonanalytic reasoning do not appear to be effective. Cancer (Cancer Cytopathol) 2013;121:329–38. © 2013 American Cancer Society.

Untutored discrimination training on paired cell images influences visual learning in cytopathology

“…In the current investigation, we start with the idea that presenting images in pairs will affect the visual learning process in naive participants, a premise that is supported by the results of our previous study . The use of paired stimuli in perceptual discrimination experiments is an accepted experimental paradigm first developed in the 1960s .…”

“…In this paradigm, images are presented only with the labels “normal” or “abnormal” and observers must learn how to extract diagnostically useful information from them, unencumbered by complex diagnostic algorithms. This “nonanalytical” approach to discrimination learning has been shown to be a rapid and effective means of visual learning in cytopathology and in other fields …”

Educational Interventions to Improve Cytology Visual Diagnostic Reasoning Measured by Eye Tracking

“…In other studies, encouraging non-analytic, pattern-recognizing (System 1) reasoning by showing multiple images has also resulted in higher diagnostic accuracy and shorter time to diagnosis. 38 Educational studies of biomedical knowledge transference have found increased transference of concepts if multiple examples are taught rather than a single example, and decreased transference when only the conceptual principle is taught. 39 In our study, group 3 students were provided with the greatest number of examples of the tumor images and, hence, an increased transference was expected.…”

From spoken narratives to domain knowledge: Mining linguistic data for medical image understanding

“…Currently, studies such as work by Evered et al [21] use non-specific measures, such as participating physicians' response time, to predict diagnostic accuracy and thus overlook much detailed information from physicians during diagnostic reasoning. We exploit human experts' knowledge to facilitate medical image grouping by applying a methodology that is more objective and automated than current research [21][22][23].…”

“…We exploit human experts' knowledge to facilitate medical image grouping by applying a methodology that is more objective and automated than current research [21][22][23]. The intuition is that the meaning of a medical image is expected to be mirrored by the spoken narrative of a physician when s/he describes the image during a diagnostic process.…”