Olfactory ERP-based classification of anosmia and normosmia using machine learning

Area of Science:

• Neuroscience
• Computational Biology
• Ophthalmology

Background:

• Olfactory dysfunction, affecting millions, lacks objective diagnostic tools.
• Current methods for assessing smell rely on subjective patient reports or basic smell identification tests.
• Developing objective measures for olfactory function is crucial for accurate diagnosis and treatment.

Purpose of the Study:

• To determine if electroencephalogram (EEG) signals can differentiate between normal (normosmia) and impaired (anosmia) sense of smell using machine learning.
• To evaluate classification performance across various single and combined olfactory stimulation conditions.

Main Methods:

• EEG data were collected from 66 participants (26 normosmic, 40 anosmic) during stimulation with phenyl ethyl alcohol (PEA), hydrogen sulfide (H₂S), and carbon dioxide (CO₂).
• Chemosensory event-related potentials (CSERPs) were analyzed using three machine learning models: support vector machine (SVM), random forest (RF), and convolutional neural network (CNN).

Main Results:

• The highest classification accuracy (86.37%) was achieved using a combination of PEA_Right + H₂S_Left + H₂S_Right stimulation.
• Convolutional Neural Networks (CNNs) generally outperformed SVM and RF models.
• Hydrogen sulfide (H₂S) stimulation demonstrated significant discriminative power, especially when presented to the left nostril.

Conclusions:

• Machine learning analysis of EEG-based CSERPs provides an effective method for distinguishing olfactory dysfunction.
• This approach demonstrates the feasibility of objective evaluation for olfactory processing.
• EEG-based machine learning offers a potential framework for data-driven diagnostic strategies for smell disorders.