Real-time AI detects EEG electrode placement with 94% accuracy

A team of researchers from Denmark has developed a real-time vision system that automatically detects and validates EEG cap electrode placement using a standard webcam. The system employs a two-stage architecture: first, a YOLOv10n single-class object detector localizes all electrodes in the image, then a geometric stage assigns each detection to a named 10-20 system position based on facial landmarks. This allows clinicians to verify correct placement instantly at the point of care.

Evaluated under strict leave-one-subject-out cross-validation across five subjects wearing clinically-validated Small/Medium/Large caps, the detector achieves a mean average precision (mAP@.5) of 0.94 with a standard deviation of 0.07. When pooling detections across all subjects, mAP rises to 0.96. The system maintains real-time throughput of 19 FPS on a commodity CPU at 640-pixel resolution, thanks to the lightweight YOLOv10n backbone. A dedicated leave-one-cap-out analysis revealed that the Small cap suffers a performance drop to mAP@.5 = 0.72, though the authors note this is confounded with subject familiarity rather than cap style.

To improve robustness, the team introduced geometric augmentation strategies—rotation, perspective transforms, and mixup—which significantly enhanced in-plane-roll tolerance and temporal-electrode recall without any inference overhead. A landmark-driven head crop extension extended the usable distance range, lifting mAP@.5 from 0.23 to 0.45 at 0.6x apparent scale. These findings suggest the system could be deployed in clinical settings to reduce manual electrode-checking time and placement errors, particularly for medium and large caps common in hospital EEG setups.