Covariance projection gave 29.1% mean site accuracy after known simulated contact loss in five SEEG sessions at 20 to 100 ms

We tested known simulated loss of one physical recording contact in public ds004457 version 1.0.2. Five patient sessions had 183 stimulation sites, 732 fit pulses, 732 test pulses, and 24 contact-loss cases. A fixed Gaussian reader used 20 to 100 ms responses. Mean site accuracy after loss was 29.1% with covariance projection and 20.0% with training mean filling. The equal-patient mean gain was 9.0 percentage points. Its descriptive patient bootstrap interval was 4.1 to 13.6 points. The exact two-sided patient test gave p = 0.125. Residual deletion gave 29.0% accuracy, so the data give little evidence for an added gain from changing the precision matrix. The before-pulse gain was 0.3 points. This exploratory offline test does not establish a population effect or live interface performance. The package preserves voltage bins, source hashes, code, all predictions and scores, figures, a Gaussian derivation, and reproduction checks. All five fresh raw extractions matched the archived inputs byte for byte.

stereoelectroencephalography · contact loss · covariance projection · Gaussian classification · electrical stimulation · offline benchmark