GeoNIRS: Geometry-Aware Pretraining for Heterogeneous fNIRS Representation Learning

Functional near-infrared spectroscopy (fNIRS) provides portable, low-cost, and spatially localized measurements of cortical hemodynamics, but substantial differences in channel count, probe layout, signal distribution, and task type hinder joint pretraining across datasets. We introduce GeoNIRS, a geometry-aware pretraining framework that uses the physical locations of source–detector pairs as a shared spatial reference across heterogeneous probe layouts. GeoNIRS compresses variable-size channel sets into a fixed-size latent representation and employs bidirectional HbO/HbR cross-reconstruction with contiguous temporal masking to learn transferable representations. GeoNIRS-base is pretrained on eight heterogeneous datasets, while GeoNIRS-large scales pretraining to 22 datasets. Neither the data nor labels of the downstream tasks are used during pretraining, enabling evaluation under a strict cross-task setting. To the best of our knowledge, GeoNIRS is the first fNIRS framework to explicitly model physical channel geometry during joint pretraining across heterogeneous channel configurations, and the first to provide both complete source code and downloadable pretrained checkpoints. We evaluate GeoNIRS on eight held-out tasks spanning clinical and auditory assessment, pain and affective processing, semantic decoding, and motor function. GeoNIRS achieves the highest task-averaged balanced accuracy among the evaluated conventional machine-learning, fNIRS-specific, general time-series, and fNIRS pretraining baselines. Compared with random initialization, pretraining improves task-averaged balanced accuracy by 4.12 percentage points under linear probing and 3.30 points under full fine-tuning, with larger benefits when downstream labels are scarce. GeoNIRS also remains robust when input channels are missing. Geometry perturbations and pretraining ablations further demonstrate that both channel geometry and cross-chromophore reconstruction contribute to the learned representations. Overall, GeoNIRS provides an open and reproducible framework for transferable representation learning across heterogeneous fNIRS datasets.