# MI-002: Optimal spatial filtering of single trial EEG during imagined hand movement > Generated from a local paper card. Do not edit this page directly; edit the source card and rerun `vp run docs:generate`. > Internal wiki boundary: paper-card evidence and user-provided PDF downloads are available for private reading. Extracted full-paper text is not published. > Source: `library/paper_cards/MI-002.md` ## Paper Access * Internal PDF: Download Paper * DOI / official page: [10.1109/86.895946](https://doi.org/10.1109/86.895946) * Deployment boundary: these PDF links are intended only for a private/protected internal wiki. ## MI-002: Optimal spatial filtering of single trial EEG during imagined hand movement ## Metadata * ID: MI-002 * Title: Optimal spatial filtering of single trial EEG during imagined hand movement * Year: 2000 * DOI / URL: 10.1109/86.895946 * Local PDF: see Paper Access section above * Text artifact: local-only path withheld from docs site * Review status: `extracted` ## Study Type * Track: MI * Task: classify left- versus right-hand motor imagery from single-trial EEG using common spatial patterns * Participants or dataset: three female right-handed subjects, age 20-27 years * Hardware: 56 Ag/AgCl EEG electrodes over central and related areas, EOG and forearm EMG monitoring * Channels or sensors: 56 EEG channels sampled at 128 Hz; artifact-contaminated trials removed; references include referential, Laplacian, bipolar, and CAR variants ## Methods * Paradigm: cued left/right hand motor imagery with randomized trials; four runs of 40 trials each * Signal processing or model: bandpass filtering in 8-30 Hz, common spatial pattern filters, variance features, linear discrimination * Training/calibration: spatial filters and classifiers estimated from subject data; best time segment selected by classification accuracy * Online/offline: offline single-trial classification analysis ## Results * Metrics: single-trial classification accuracy * Main findings: best classification results for the three subjects were 90.8%, 92.7%, and 99.7%; more than two channels improved performance, while 56 channels did not significantly improve over an 18-channel subset around hand areas * Reported limitations: artifact-free data needed for reliable filter calculation; electrode position changes can reduce CSP benefit; CSP may require many electrodes; only variance features are used ## Relevance To This Project * Supports: CSP as a foundational MI spatial-filtering baseline and the importance of motor-area channel placement * Conflicts with: offline binary hand imagery does not establish online confirm/cancel/pause/no-control reliability * Design implication: Exp2 should treat CSP/FBCSP as baseline MI decoders and explicitly test no-control false activation and electrode-set robustness ## Extracted Evidence | Claim | Status | Evidence Note | Page/Section | | --- | --- | --- | --- | | CSP can extract discriminative spatial information from multichannel EEG during left/right motor imagery. | verified | The abstract and methods describe CSP filters estimated from two populations of single-trial EEG for left and right imagery. | Abstract; Methods | | High offline left/right MI accuracy was achieved in three trained subjects. | verified | The paper reports best classification results of 90.8%, 92.7%, and 99.7%. | Abstract; Results | | Motor-area electrode coverage matters for MI classification. | verified | Results indicate that 18 channels covering hand areas were sufficient for good classification and improved over two channels in some subjects. | Results; Discussion | | CSP performance can be sensitive to artifacts and electrode placement. | verified | The discussion states that artifacts can affect filter estimation and that changing electrode positions may reduce classification improvements. | Discussion | | SAH-BRI-Grasp can use CSP as an MI decoder baseline, but must validate online no-control behavior. | inferred | The paper is offline binary MI and does not evaluate mode-control commands or robot safety. | Methods; Discussion | ## Open Questions * Which motor-area channel set will be available in the project hardware? * Should Exp2 compare CSP, FBCSP, EEGNet, and Riemannian baselines under the same no-control protocol? * How strict should artifact rejection or online quality gating be before accepting MI commands?