GlobeMetrics: A Healthcare Framework for Video Based Saccade Characterization

Eye movement analysis is extensively utilized in understanding mechanisms governing perception, cognition, and action and proves valuable in exploring neurological and neurodegenerative diseases. This paper introduces GlobeMetrics, a healthcare application designed for ocular analysis from video data. Our emphasis in this study is on estimating the saccadic profile of the subjects utilizing the GlobeMetrics framework. The framework includes a setup for data recording, an appearance-based gaze estimation system, a module for analyzing gaze data specifically for saccade analysis, and an interactive GUI application that interfaces with each aspect of the framework. The proposed gaze estimation network consists of a convolutional neural network (CNN) based segmentation and regression network that maps input frames to gaze points. The proposed gaze estimation architecture achieves a prediction error of 0.467± 0.133 cm on our database. Additionally, the segmentation network attains mean IOUs of 95.19 and 97.39 for sclera and iris, respectively. Our proposed framework, GlobeMetrics, offers an interactive platform for conducting ocular analysis in clinical settings. This application seamlessly integrates data recording, stimulus generation, database management, and data analysis within a unified framework. The overall framework is accurate, robust, and generalizes well to new subjects.