Avalilação PREreview de Brain Control of a Computer Cursor for Online Target Selection - A Non-Invasive BCI for Continuous Movement Decoding

Peer Review Report

By Folarera Kassim (ORCID iD: 0009-0007-2148-5675)

Overview and Scope

The manuscript presents a non-invasive brain-computer interface framework for continuous cursor control using electroencephalography data. It addresses a major operational challenge by incorporating an intentional stationary state alongside active navigational vectors. Users complete target acquisition tasks by sustaining gaze or intention over a fixed area to trigger selection validation. Because the system does not require residual muscle activation, it may offer meaningful utility for individuals with total motor paralysis.

Core Conceptual Strengths

The design addresses a longstanding BCI problem in which every user action or gaze path is incorrectly treated as an active command. By introducing a distinct resting state, the system allows users to process visual information without causing unintended cursor drift across the interface. Training the continuous decoder without relying on physical movement imagery also broadens the clinical accessibility of the device. The empirical testing suggests that participants can reliably transition between navigational states and achieve performance above chance.

Major Methodological and Theoretical Flaws

The study’s most significant limitation is the complete absence of processing latency calculations. The delay introduced by EEG buffering, feature extraction, and spatial filtering windows is not quantified. This unaddressed delay creates a feedback gap that may destabilize real-time error correction.

This latency bottleneck likely contributes to the very low information transfer rate, as participants averaged only seven selections out of forty attempts. Delayed visual feedback appears to force users to overcorrect their motor intent, producing spatial oscillations and cursor overshooting near target boundaries. The authors do not provide an analytical model of these dynamics, nor do they report standard information transfer metrics in bits per minute.

The architecture also lacks a formal credit assignment model to explain how the brain processes delayed success signals during online training. This limitation is reflected in the low performance ceiling, with even experienced BCI users achieving fewer than thirteen successful target selections.

Statistical and Data Presentation Deficiencies

The empirical validation is weakened by unclear statistical reporting regarding the significance baseline. The authors state a threshold of four and a half targets but do not explain whether it was derived using Monte Carlo permutation testing or an exact binomial distribution.

The presentation of the data would also benefit from spatial trajectory plots. These missing diagrams are necessary to determine whether trial failures were caused by directional decoding errors or an inability to maintain the stationary state. In addition, the manuscript does not provide training matrix data showing the experience thresholds or practice hours used to distinguish naive subjects from expert users.

Minor Clarity and Structural Revisions

Several passages would benefit from tighter phrasing and clearer transitions. In particular, the manuscript should define technical terms more precisely, avoid repeating similar claims about control accuracy, and clarify how performance metrics were calculated. A brief statement summarizing the practical implications of the framework would also improve the overall structure.

Overall Assessment

The manuscript offers an interesting and potentially important approach to non-invasive BCI control, especially for users with severe motor impairment. However, the current version would be substantially strengthened by a clearer treatment of latency, more transparent statistical reporting, and more complete data presentation.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.