Publication

Abstract : In this research, we delve into the predictive capabilities of deep learning models for classifying Areas of Interest (AOIs) based on gaze patterns in patient assistance systems. Utilizing eye-tracking technology to capture gaze data consisting of x and y coordinates along with duration, we aim to discern the specific regions of interest on the screen. With 9 distinct AOIs identified through a 9-point calibration process, we leveraged Python to preprocess the dataset. Applying FFNN, LSTM, and GRU, EyeHelp achieved promising accuracies: 80% for FFNN, 96% for LSTM, and 93% for GRU. Comprehensive observations in terms of accuracy, recall, and F1-score metrics across AOI classes are provided. EyeHelp employs an open source integrated webcam; hence the approach is open and adaptable to various healthcare environments. By integrating this technology, our system can be seamlessly deployed in clinics, hospitals, and even home care environments. The utilization of integrated webcams facilitates cost-effective and non-intrusive monitoring, offering valuable insights into patient behavior and attention patterns.