학술논문
Incorporating flash adjacency into the classifier for a language model-based P300 Speller
Document Type
Conference
Author
Source
2024 IEEE 4th International Conference on Human-Machine Systems (ICHMS) Human-Machine Systems (ICHMS), 2024 IEEE 4th International Conference on. :1-8 May, 2024
Subject
Language
Abstract
The P300 speller is a common brain-computer interface (BCI) application designed to allow patients with neuromuscular disorders such as amyotrophic lateral sclerosis (ALS) produce text output through the detection of P300 signals in their electroencephalogram (EEG) signals. The standard P300 speller relies on the detection of signals evoked by visual stimuli, usually consisting of rows and columns highlighted in a grid of characters. Since the visual field is substantially larger than these stimuli, adjacent flashes to the attended characters may cause false positive signals and lead to erroneous output. Previous studies have tried to address this issue by limiting the number of adjacent stimuli. However, it is not possible to completely avoid adjacent stimuli, so the problem cannot be eliminated. In this study, we instead account for these adjacency false positives in the classifier by utilizing information from adjacent flashes to optimize the system. This optimization is accomplished by adding a bias to the target character detection based on adjacent flashes which creates a new probability model to improve the accuracy and speed of classification. We tested our adjacency classifier in both the standard P300 paradigm and in conjunction with natural language processing. The new algorithm was evaluated offline on a dataset of 69 healthy subjects, which showed increases in speed and accuracy when compared to standard classification methods. On a population level, our adjacency model led to increased performance in the standard P300 paradigm, but the improvement in the presence of NLP was not significant. On an individual level, some subjects had substantial improvements in both settings (the number of subjects who saw an ITR increase of at least 5 bits/minute were 57.4% and 21.7% for the standard and NLP paradigms, respectively), suggesting that incorporating adjacent flesh information into the classifier can potentially provide a better communication system for some users.