A repeat difference metric improved matching of 165 sounds in six SEEG patients and 24 fMRI adults

We tested whether spatial correlations help identify repeated sound responses in three separate human studies. SEEG supplied high-gamma responses. Two fMRI studies supplied BOLD estimates. Training sounds set site scales, offsets and distance rules. Other sounds tested matches within broad source groups. Development used two SEEG patients and four fMRI adults. A local code lock preceded tests in six other patients and 24 other adults. All 30 reserved people improved with the repeat difference metric against both a site variance rule and a mean response correlation rule. Mean gains over the site variance rule were 1.446, 1.302 and 4.833 AUC percentage points in SEEG, B2021 and NH2015. Gains over the response rule were 1.470, 5.717 and 3.150 points. Both fixed shared claims passed Holm correction, p=0.03125. Each required positive mean confidence limits and positive-person sign tests in every cohort. An exploratory third comparison found that the response rule hurt B2021 but helped NH2015. An unfitted Gaussian model had perfect covariance map agreement but different matching scores. Thus map agreement alone did not justify a substitute metric or an optimal decoder. The package gives figures, person uncertainty, all outcomes, code, source pins and repeat instructions. Sparse clinical placement, source selection, different ages, shared sounds and unequal filters limit scope. This is a fixed-sound measurement benchmark. No common neural cause is established.

human auditory cortex · SEEG · fMRI · covariance estimation · sound identification · noise normalization · cross-modal test