Toward Plug-and-Play P300-Based BCIs: Resolving Key Challenges with Single-Trial, Calibration-Free Systems, and Task-Relevant Stimulus Design

Abstract

Brain-computer interface (BCI) systems based on the P300 event-related potential (ERP) have shown great promise for enabling non-invasive communication and control. However, several challenges remain, including the necessity for repeated stimulus presentations, individual calibration, and variability in user performance. Addressing these challenges is critical to advancing BCIs toward real-world, plug-and-play usability. In this study, we present and validate a novel P300 paradigm design of task-relevant dynamic stimulus (based on finger-tapping animation) together with developed optimal computational models to increase the performance of P300-based BCI systems. We demonstrate improvements in accuracy, information transfer rate (ITR), and usability over a range of settings from single-trial to calibration-free paradigms using a large number of online and offline experiments. The core of this study is the development and application of the finger-tapping stimulus, which dynamically replicates a task-relevant motor action and captures selective attention in the oddball paradigm for P300-based BCIs. In initial online experiments involving 37 healthy participants, we assessed six stimulus-task combinations comprising three types of stimuli—color-changing, icon- rotating, and finger-tapping—paired with two mental tasks: counting and motor imagery (MI). Results revealed that the finger-tapping stimulus consistently outperformed conventional stimuli, achieving an average online accuracy of 91.17% and an ITR of 28.37 bits/min with two stimulus repetitions when paired with the counting task. This marks an 11.17% improvement over the lowest-performing paradigm (color-changing stimulus with counting). These findings highlight that the presentation of more task-relevant and dynamic stimuli can enhance ERPs with consequential improvements in BCI performance. Offline analyses were conducted to demonstrate the single-trial and calibration-free P300-based BCI feasibility so that we can more deeply refine and expand the system. Offline performance showed the single-trial feasibility of the developed paradigm, attaining 91.7% accuracy and an ITR of 59.92 bit/min. Additionally, across-subject calibration-free models were tested, where a BCI trained on data from 36 participants was directly applied to a left-out participant without individual calibration. This configuration also yielded high performance, with 87.75% accuracy and an ITR of 52.61 bits/min at the single-trial level. The finger-tapping stimulus proves to be stable and can readily replace the labor- intensive individual calibration, serving as an important step toward a plug-and-play BCI system with proper computational methods. To verify system performance in real-world conditions, we conducted several online experiments based on these offline findings. This phase involved both full-channel (31ch) and reduced-channel (8ch) configurations, representing high-resolution laboratory setups and portable, cost-effective implementations, respectively. Under single-trial, calibration-free conditions with the 31-channel setup, the system achieved an average accuracy of 85.75% and an ITR of 46.42 bits/min. A single-trial, calibration condition achieved a high accuracy of 91.67% and an ITR of 57.35 bits/min utilizing the 8 channels. The system achieved a respectable 78.89% accuracy and an ITR of 36.49 bits/min even in a single-trial, calibration-free configuration at 8 channels. These results demonstrate the scalability and the adaptability of our approach, which achieves a high performance independently on hardware configurations and experimental paradigms. Our ERP analysis provided critical insights into the mechanisms that explain why finger-tapping was more effective for enhancing performance. In the N200 component, finger-tapping showed shorter latencies than both color-changing and icon-rotating stimuli at Oz (p = 6.699 × 10−8). The latency reduction indicates that the finger-tapping stimulus leads to more rapid cognitive processing and attentional engagement. This reduction in latency suggests that the finger-tapping stimulus facilitates faster cognitive processing and attentional engagement. Although the N200 peak amplitude did not show significant differences across stimulus types, the latency advantage highlights the stimulus’s effectiveness in capturing attention and expediting neural processing. Spatial effects of the finger-tapping stimulus were evident in the P300 component. The finger- tapping stimulus elicited smaller amplitudes than the icon-rotating stimulus at Cz. At posterior locations especially Pz and Oz, finger-tapping stimulus elicited larger than conventional P300 amplitudes and shorter latencies. For example, at Oz, the P300 amplitude for the finger-tapping stimulus was larger than that of both the color-changing (p = 5.8796×10−8) and icon-rotating stimuli (p = 5.18×10−8). The increases in posterior sites are suggestive of increased visual attention and processing that would be necessary for proper ERP elicitation. Moreover, the finger-tapping stimulus was associated with shorter P300 latencies recorded at posterior channels than the other stimuli, thus further confirming its capacity to foster rapid and strong neural responses. Support vector machine (SVM) weight vector analysis corroborated these ERP findings, identifying parietal and occipital regions (including Pz and Oz) as critical for accurate classification. The ERP improvement and high SVM weights correlation in these areas demonstrate the neural and computational synergy achieved by the finger-tapping stimulus. These findings collectively demonstrate that task-relevant dynamic stimuli can shift neural activation patterns to posterior regions, enhancing the overall reliability and efficiency of BCI systems. Notably, this finger-tapping stimulation reduced the variability with which participants performed thus making both P300-based BCIs less difficult but also functioning with greater stability. The finger- tapping design yielded a massive increase for participants who fared poorly with conventional stimuli. This performance difference reduction indicates that such task-specific and natural stimuli can adapt to a wide range of user profiles, thereby expanding the accessibility and usability of P300-based BCIs. This study marks a significant step forward in the development of plug-and-play P300-based BCIs. We demonstrate, through a careful design of a simple task-related stimulus and state-of-the-art machine learning methods, that two major obstacles to practical implementation (the need for individual calibration and multiple repetitions of the same purely physical stimulus) can be overcome. The proposed system achieves high accuracy, rapid information transfer, and reduced user variability, even under single-trial and calibration-free conditions. The scalability demonstrated across both high- resolution and reduced-channel configurations reinforces the system’s adaptability to diverse applications, from laboratory research to portable consumer-grade devices.