TY - GEN
T1 - Maintaining High Accuracy General P300 Speller Using the Language Modeling and Dynamic Stopping
AU - Soetedjo, James
AU - Keluo-Udeke, Osita Sean
AU - Amold, Corey
AU - Pouratian, Nader
AU - Speier, William
N1 - Publisher Copyright:
© 2020 IEEE.
PY - 2020/10
Y1 - 2020/10
N2 - Patients with neuromuscular diseases such as amyotrophic lateral sclerosis can have difficulty communicating because neural signals cannot reach effector muscles. Recent advances in brain-computer interfaces have allowed these patients to communicate by converting neurological signals into computer commands. One common brain-computer interface is the P300 speller, a system that allows these patients to spell out text. Because of the electroencephalogram (EEG) signal variability between patients, it is hard to create a classifier applicable to all patients. Therefore, current methods use an arduous training step personalized for each patient. There have been previous attempts to create a general classifier that works for all subjects, but these attempts have generally resulted in poor accuracies that were insufficient for practical use. This paper presents a novel cross-subject approach for the P300 speller. It uses a language model which adjusts the probabilities of each character based on context to improve classifier performance. Additionally, dynamic stopping allows the system to continually obtain EEG signal from the patient until the system is confident in its character selection. By using these two approaches, we can maintain reasonable selection accuracy, allowing subjects to use the system without an individualized training step.
AB - Patients with neuromuscular diseases such as amyotrophic lateral sclerosis can have difficulty communicating because neural signals cannot reach effector muscles. Recent advances in brain-computer interfaces have allowed these patients to communicate by converting neurological signals into computer commands. One common brain-computer interface is the P300 speller, a system that allows these patients to spell out text. Because of the electroencephalogram (EEG) signal variability between patients, it is hard to create a classifier applicable to all patients. Therefore, current methods use an arduous training step personalized for each patient. There have been previous attempts to create a general classifier that works for all subjects, but these attempts have generally resulted in poor accuracies that were insufficient for practical use. This paper presents a novel cross-subject approach for the P300 speller. It uses a language model which adjusts the probabilities of each character based on context to improve classifier performance. Additionally, dynamic stopping allows the system to continually obtain EEG signal from the patient until the system is confident in its character selection. By using these two approaches, we can maintain reasonable selection accuracy, allowing subjects to use the system without an individualized training step.
KW - P300 Speller
KW - brain-computer interface
KW - classifiers
KW - language model
UR - http://www.scopus.com/inward/record.url?scp=85099565417&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85099565417&partnerID=8YFLogxK
U2 - 10.1109/BIBE50027.2020.00066
DO - 10.1109/BIBE50027.2020.00066
M3 - Conference contribution
AN - SCOPUS:85099565417
T3 - Proceedings - IEEE 20th International Conference on Bioinformatics and Bioengineering, BIBE 2020
SP - 365
EP - 368
BT - Proceedings - IEEE 20th International Conference on Bioinformatics and Bioengineering, BIBE 2020
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 20th IEEE International Conference on Bioinformatics and Bioengineering, BIBE 2020
Y2 - 26 October 2020 through 28 October 2020
ER -