Added guide patterns reduced test sound response error in SEEG and two fMRI cohorts with source processing

We tested whether patterns at eight guide sites predict responses at eight other sites to sounds omitted from fitting. Public human SEEG high-gamma activity and BOLD fMRI came from separate studies. We used 128 source-selected sounds, four sound folds and different repeats for guides and targets. A baseline used the first guide principal component. The candidate kept that prediction and added the other seven components. Development used two patients and four adults. A local code lock preceded tests in six other patients and 16 other B2021 adults. Mean gains were 2.304% and 5.661% of standardized target energy. Their 95% person t intervals were [1.123, 3.485]% and [4.700, 6.622]%. All 22 scores were positive. The fixed mean-and-sign rule passed; the largest component p was 0.015625. A separate NH2015 study gave 2.673% gain in eight adults, interval [1.215, 4.131]%, with all eight positive. A short-baseline subtraction check gave SEEG mean -0.291%, interval [-0.399, -0.182]%, with no positive patient. Same-repeat scores were larger. Fine-category removal left an uncertain gain in the external fMRI study. A constructed one-source model gave a large gain with independent repeat errors. Thus the result concerns prediction in the source-processed records. It does not establish several neural generators, common anatomy or causality. Independent fits checked all 36 development and test people within 1.4e-15. The package gives code, all results, source pins and terms, local plans, figures and repeat instructions. Original neural arrays stay external.

auditory cortex · SEEG · fMRI · response prediction · sound generalization · principal components · measurement noise · reproducibility