High performance computation of human computer interface for neurodegenerative individuals using eye movements and deep learning technique

• Published in: The Journal of supercomputing Vol. 78; no. 2; pp. 2772 - 2792
• Main Authors: Ramakrishnan, Jayabrabu, Doss, Rajesh, Palaniswamy, Thangam, Faqihi, Raddad, Fathima, Dowlath, Srinivasan, Karthik
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.02.2022; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial neural networks; Compilers; Computer Science; Cornea; Deep learning; Eye movements; Human performance; Interpreters; Machine learning; Mobile and Intelligent Sensing on High Performance Computing; People with disabilities; Processor Architectures; Programming Languages; Hilbert Huang transform; Electrooculography; Pattern recognition neural network; Human–computer interface; Feedforward neural network; Periodogram; Ant lion optimization
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-021-03932-z

Disabilities due to neurodegenerative disease are rapidly increasing in number. The need for rehabilitative devices to achieve a normal and comfortable life in the absence of biochannels has also increased. The activities of biochannels can be easily replaced by implementing rehabilitative devices. The electrooculography (EOG)-based human–computer interface (HCI) is one of the most important techniques for enabling disabled persons to enjoy a normal life. The technique of measuring the cornea-retina potential difference is called EOG. The technology for converting thoughts to different control patterns to activate external devices is called the HCI. In this paper, we carried out a study on ten subjects aged 20–30 years using a five-electrode signal acquisition system (AD T26). The subject performances were experimentally verified by implementing periodogram features with a feedforward neural network trained on a nature-inspired algorithm. The analysis was conducted offline and online to evaluate the achievement of the developed HCI. The study showed an average classification accuracy of 93.93% for four tasks, with 95% accuracy for the offline mode and 90.12% for the online mode. The participating subjects drove the mobile robot in all directions frequently and quickly with a recognition accuracy of 98.12%. The study confirmed that the four tasks (related to driving the external device) performed by the subjects were convenient to perform in real time.