iMINDBench: iEEG Multi-Institution Neural Decoding Benchmark

Intracranial electroencephalography (iEEG) is widely used to record electrical activity directly from electrodes inside the human brain, making it an attractive modality for neural decoding. However, progress in iEEG decoding remains difficult to measure reliably: datasets are task- and/or institution-specific, preprocessing choices strongly influence performance, and evaluation protocols often differ across studies. We introduce iMINDBench, a multi-institution benchmark for naturalistic movie-watching iEEG decoding built from independently collected movie-watching datasets. iMINDBench aligns fifteen language, auditory, and visual decoding tasks across three datasets, fixes target evaluation splits, and reports performance across data-scaling regimes and unit decodability subsets. Within each preprocessing track, pretrained models improve over non-pretrained baselines, yet strong Multi-STFT baselines remain competitive across tracks. In our scaling study, adding up to 25 times more supervised data from other subjects or institutions yields only small or task-dependent gains over within-session training. These results show that current gains do not yet translate into reliable cross-subject and cross-institution scaling.