A joint team from KTH Royal Institute of Technology and Karolinska Institute demonstrates that olfactory brain–computer interfaces can detect odor perception from single trials using electrobulbogram and EEG signals processed with ResNet-1D convolutional neural networks, marking a milestone in non-invasive sensory BCI technology.
Key points
A ResNet-1D CNN achieves significant above-chance AUC-ROC for scalp-EBG (t=4.15), EEG (t=5.29), and source-EBG (t=3.21), confirming single-trial odor detection feasibility.
Four-electrode electrobulbogram (EBG) on the forehead matches 64-channel EEG performance for olfactory signal classification, enabling simpler hardware setups.
Fusing scalp-EBG with sniff-trace data improves logistic regression detection (t=2.70, p=0.009), demonstrating multimodal synergy between brain and respiratory signals.
Why it matters:
This study pioneers single-trial olfactory BCI detection, laying groundwork for sensory-enhanced human–machine interfaces beyond traditional visual and motor modalities.
Q&A
What is an electrobulbogram (EBG)?
Why is single-trial odor classification challenging in EEG?
How does ResNet-1D process brain signals?
What does AUC-ROC indicate in classification?
Read full article
Academy
Olfactory Brain–Computer Interfaces
Introduction
Olfactory brain–computer interfaces (BCIs) aim to decode smell perception directly from neural signals. Unlike visual or motor BCIs, olfactory BCIs must capture activity from deep brain structures such as the olfactory bulb, which is the first relay station for odor information. Recent advances in electrode design and signal processing have enabled non-invasive recordings that reveal odor-evoked neural patterns.
The Olfactory System
The olfactory system begins in the nasal epithelium, where odorant molecules bind receptors on sensory neurons. These neurons relay signals to the olfactory bulb, generating oscillatory activity in beta (10–30 Hz) and gamma (30–70 Hz) bands. The only primary sensory system that projects directly to cortical areas, olfaction bypasses the thalamus, making real-time decoding more challenging.
- Olfactory Bulb: Processes initial odor signals; activity can be recorded via electrobulbogram (EBG).
- Piriform Cortex: Involved in higher-order odor discrimination and memory.
- Sniff Trace: Breathing patterns provide complementary information on inhalation timing.
Electrobulbogram (EBG)
EBG uses a compact array of four electrodes on the forehead to capture bulb activity. By focusing on the odor-responsive region, EBG improves the signal-to-noise ratio compared to standard 64-channel EEG, simplifying hardware requirements for portable devices.
Signal Processing
Decoding odor perception from single trials uses time–frequency analysis. Morlet wavelets transform raw signals into spectrograms, emphasizing oscillatory power changes in beta and gamma bands between 0–1 s after sniff onset. These representations feed into machine learning classifiers.
Deep Learning with ResNet-1D
ResNet-1D is a one-dimensional residual convolutional neural network tailored for temporal data. Skip connections stabilize training on limited biomedical datasets, automatically extracting relevant features and mitigating vanishing gradients.
Challenges and Future Directions
Major hurdles include low signal-to-noise ratio in single trials, high inter-subject variability, and limited data due to long inter-stimulus intervals. Future work focuses on large-scale data collection, advanced data augmentation, and transfer learning to improve generalization and reliability in real-world olfactory BCI applications.
