Kaiwen Mo
Research Mentor(s): Jane Huggins
Mentor Department: Physical Medicine and Rehabilitation
Authors: Kaiwen Mo, Jane Huggins
Session: Session 3 (11:00am – 11:50am)
Presentation Type: Poster 49
Abstract
This study explores the calibration of a brain-computer interface (BCI), a system that interprets brain signals to control external devices, focusing on aiding individuals and children with limited attention spans due to severe motor or speech impairments. Traditional BCI systems require active user engagement for calibration, creating barriers for those lacking established choice-making skills. Our objective is to gather high-quality calibration data characterized by clear and consistent patterns of brain activity that indicate user interest and engagement. We do this by presenting participants with positive stimuli, including images and their scrambled versions, and rewarding them with video clips if they choose the original images. Calibration uses machine learning to map brain signals to choices by assigning weights to specific neural responses. This process generates a score that reflects user engagement. By analyzing these scores, the BCI can effectively determine user preferences, enhancing communication for users unable to provide direct responses. Our primary aim is to develop a calibration methodology that effectively captures participants’ attention and focuses their brain activity on the stimuli, ensuring the BCI can provide reliable communication tools tailored to individual needs. We are currently sourcing video clips from the copyright-free website Pexels, focusing on topics such as animals, food, and vehicles, with 30 videos per category, intended to evoke positive responses. Screenshots manually extracted from these videos are adjusted in 480 x 270 format, and are used to create two distinct scrambled versions via a custom MATLAB function. Once all images are collected and scrambled, we create the parameter files used by the executable program, which adjusts settings, and displays the images and scrambled versions to participants, followed by videos as a reward. We create the parameter files by using a custom MATLAB function that reads the directory of the folders and loads the parameter files for each of the categories. Our efforts have resulted in collecting 30 video clips for each of the 3 categories. We have started integrating these images into the parameter files, beginning with the food category. We also need to consider that participant interest in stimuli varies, with some showing less interest in the available categories. This highlights the need to diversify stimuli to better match individual preferences, thereby increasing participant engagement. We anticipate having 3-5 categories of videos and images, each accompanied by parameter files that can properly display the images and reward participants with video clips when they select the original image over the scrambled one. This research aims to develop BCIs that users can voluntarily engage with, offering calibration that requires minimal active effort. The parameter files developed here could serve as foundational tools for participant tests that contribute to more advanced future projects. These insights are crucial for refining the stimuli selection process and ensuring the data collected will support the development of personalized BCI calibration profiles. By the end of the semester, the goal is to deepen our understanding of BCI parameter files and coding.
