Eye-tracking as a proxy for coherence and complexity of texts, and for the interpretation of cytopathological images

Abstract

Eye movements offer a unique insight into cognitive processes, bridging the mechanics of vision with underlying neural activities. Through swift movements called “saccades”, our eyes direct attention to areas of interest, while “fixations” provide moments for detailed visual information gathering. In tasks like reading, fixation patterns reveal cognitive demands and processing depth. Eye tracking, a non-invasive method, captures these movements with precision, enabling the study of visual behavior across tasks and stimuli complexities. By analyzing eye movement data using statistical physics and machine learning tools, we gain insights into collective behaviors and cognitive processing patterns. Equipped with a cutting-edge eye tracker, our research focused on eye-tracking experiments across various visual tasks, resulting in two main projects published in reputable journals. Our primary project, “Eye-tracking as a Proxy for Text Complexity and Coherence”, involved experiments where participants read diverse texts. Using a binary model for fixation sequences akin to an Ising system integrated into the Maximum Entropy Model, we examined fixation patterns. Leveraging results from an extensive survey, we compared these findings with properties derived from eye-tracking experiments. Our analysis revealed a positive correlation between text complexity and average fixation density, measured by magnetizations. Additionally, statistical modeling and Pairwise Maximum-Entropy analysis demonstrated the effectiveness of critical point distance in assessing text coherence, indicating a potential for robust cognitive measures compared to conventional survey responses. In parallel, a collaborative project with the ``Engenheria em Teleinformatica'' post-graduate program focused on visual attention among cytopathologists analyzing Pap smear slides. This research offered insights into visual mechanisms during Pap smear analysis, potentially enhancing current practices and informing the development of innovative analysis techniques. Our investigation encompassed a review of visual attention dynamics, exploration of saliency learning models for medical diagnosis, and evaluation of eye-tracking applications in medical imaging. Through the comparison of eye fixation maps with saliency prediction algorithms, the reliability of these algorithms in detecting clinically relevant cells was assessed. The findings from our primary project demonstrate the robustness of the maximum-entropy approach for probing diverse complex systems effectively. Additionally, our observations highlight how humans respond cohesively to coherent textual information, reflecting advanced language formation and reading prediction mechanisms. Moreover, our secondary project revealed a significant agreement in gaze patterns among participants, supporting the identification of clinical Regions of Interest in Pap smear images. Further investigation into saliency models provided valuable insights into their predictive accuracy and potential implications for medical image interpretation. Lastly, our interdisciplinary research, spanning neuroscience, linguistics, and computer science, provided valuable insights into applying physical models to study cognitive activity.