Microscopy is an important skill with which to acquire knowledge for a basic understanding of biology and medicine. Working with a microscope is a skill that needs to be acquired and that develops with training and experience. Assessing someone’s level of expertise in this domain is crucial for at least two reasons: certification and educational purposes. The standard method in this regard is to ask a subject specific questions about a given histological specimen that he/she inspects under the microscope. However, from the perspective of Cognitive Science this method must be considered an off-line measure, because it measures the outcome of what is actually a conglomerate of processes, i.e., visual search, object identification, and information retrieval from long-term memory.
The main goal of this project is to use eye tracking as an on-line method for competence assessment in microscopy. This idea rests on the long established fact that eye movements directly reflect what humans attend to and for how long (eye-mind-link), which in turn should allow to draw conclusion about all cognitive processes involved in microscopy. Given that generally experience influences cognitive processing in many different domains, it is reasonable to expect specific gaze allocation patterns depending on a microscope user’s expertise.
The first step in this project is to establish a correlation between the existing (to be optimized) off-line and the to be developed (and to be optimized) on-line measures (eye movement data). Although, in this regard some efforts have been made previously, reliably distinctive patterns from visually processing a histological specimen have not emerged for novices and experts, so far. Most likely this is due to special characteristics of eye tracking data.
Eye tracking data have a specific form that makes their statistical analysis quite interesting and appealing. The raw data are gaze points of a participant looking at a stimulus. The temporal development of these points can be summarized in fixations and saccades. There are several ways to represent these objects mathematically. One can identify areas of interest of the stimulus on which fixations are focused. Possible summary statistics are the lengths of fixations and the numbers of transitions between areas of interest. Currently, statistics of eye tracking data is not well developed. There exist many proposals for visualization of features of eye tracking data and for the use of the visualizations as data-analytic tool. But it is difficult to base a rigorous statistical analysis on this data-analytic tools. On the other side, statistical methods have been used only for the analysis of simple summary statistics. E.g. parametric fits have been proposed for the probability to focus one specified area of interest. This can for instance be done by using logistic regression. But it may be argued that too much information gets lost when the eye tracking data are reduced to this low-dimensional summary statistics. Furthermore, the understanding of the predictor variables may have been too simplistic in previous research. To carefully evaluate the correlation between on-line and off-line measures it may be necessary to consider various subject-related variables, which may contribute differently to what is currently subsumed under terms such as ’novice’, ’intermediate’, or ’expert’.
Once it has been shown that viewing behavior equally well, or better allows to assess a microscope user’s skills, the second step in this project is to automate the assessment. Ideally, the tool to be developed here automatically analyzes the viewing behavior of a given candidate, delivers a score (or a series of scores) that reflects the level of performance, and a measure that captures the reliability of the analysis. All of this may be achieved by comparing a candidate’s data with one or several baselines.