.png){kind=icon}
.png){kind=icon}
Return
Single-Trial Real-Time-Compatible P300 Latency Estimation Using a Deployment-Consistent Deep Learning Regression Framework
.png){kind=icon}
OA
AI
DOI：10.1109/access.2026.3716564.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate single-trial estimation of P300 latency remains challenging due to low signal-to-noise ratio, substantial trial-to-trial variability, and the limitations of conventional peak-based ERP analysis. This study presents a deployment-consistent deep learning framework for real-time single-trial P300 latency estimation, combining Center of Mass (COM)-based latency labeling, image-based signal representation, and sequential streaming inference. Rather than focusing solely on offline prediction accuracy, the proposed framework is designed to support real-time-compatible latency estimation under practical deployment conditions. A total of 11,057 single-trial Pz waveforms obtained from 122 subjects were transformed into standardized image representations and paired with COM-derived latency labels. The dataset was partitioned into training (N = 7,740), validation (N = 1,106), and independent test (N = 2,211) subsets. Three baseline approaches were implemented for comparison: conventional peak detection, template matching, and a direct 1D CNN regression model operating on the raw waveform. The proposed COM-based 2D CNN achieved the best overall performance, yielding a mean absolute error (MAE) of 5.49 ms, a root mean squared error (RMSE) of 7.12 ms, and an R2 of 0.8535 on the independent test set. Conventional peak detection and template matching produced negative R2 values, whereas the 1D CNN baseline achieved an R2 of 0.7604. During deployment-consistent sequential streaming evaluation, identical performance was maintained without retraining or parameter updates, demonstrating real-time applicability. These findings indicate that COM-based labeling and image-based deep regression provide a robust framework for single-trial P300 latency estimation while eliminating the need for trial averaging, handcrafted feature extraction, and manual latency selection.
Keywords:
P300 latency estimation
single-trial EEG
center of mass (COM)
deep learning regression
streaming inference
real-time EEG analysis
Journal
IF:
3.6
Papers:
9.7W
Citations:
29.4W