Machine learning-based feature combination analysis for odor-dependent hemodynamic responses of rat olfactory bulb

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Rodents possess a keen sense of smell vital for detecting various substances.
• Current methods using rodents for detection are limited by inconsistent concentration.
• Brain-computer interface (BCI) technology offers a potential solution for inferring detected odors without behavioral cues.

Purpose of the Study:

• To evaluate the efficacy of non-invasive near-infrared spectroscopy (NIRS)-based BCI for odor detection and identification in rats.
• To optimize machine learning models by extracting and analyzing features from hemodynamic responses.
• To identify key features for accurate odor inference.

Main Methods:

• Utilized NIRS-based BCI technology, a non-invasive approach measuring neuronal activity.
• Extracted six hemodynamic response features: slopes, peak, variance, mean, kurtosis, and skewness.
• Analyzed the importance of individual and combined features for machine learning models.

Main Results:

• NIRS-based BCI technology successfully detected and identified odors from the rat olfactory system.
• The 'slopes' feature demonstrated the highest F1-Score for odor inference.
• A combination of 'slopes' and 'mean' features proved most effective for odor inference.

Conclusions:

• Non-invasive NIRS-based BCI is a viable method for olfactory odor detection and identification in rats.
• Hemodynamic response feature engineering, particularly 'slopes' and 'mean', is critical for developing accurate bionic nose systems.
• This research provides a foundation for future advancements in hemodynamic response-based bionic nose technology.