New AI Model Reads Brain Activity With Fewer Sensors
This could make brain-computer interfaces cheaper and more practical for everyday use.
The article reports a fNIRS foundation model called MtReNIRS, a montage-aware masked autoencoder pretrained on 12 public fNIRS datasets spanning 268 subjects and 567 recording hours. It learns generalizable neurovascular representations that transfer across three external datasets by jointly encoding fNIRS’s slow hemodynamic nature, chromophore identity, and cortical location and mapping heterogeneous montages onto a shared anatomical reference.
MtReNIRS achieves cross-subject performance on par or better than fully supervised baselines, excels in few-shot within-subject calibration outperforming prior state-of-the-art fNIRS models, and remains robust to sparse montages. The authors say this montage-aware self-supervision provides a scalable foundation for label-efficient brain-computer interfaces.
- MtReNIRS is an AI model that interprets brain activity from fNIRS, a cheap and portable brain-scanning method.
- It was trained on 12 datasets with 268 subjects and 567 hours of data, and can adapt to new users with minimal calibration.
- This could lead to more affordable and practical brain-computer interfaces for medical and consumer use.
Why It Matters
This could make brain-computer interfaces cheaper and easier to use, helping people with disabilities and enabling new consumer applications.