Parkinson’s disease (PD) is a progressive neurodegenerative disorder that impairs motor and sensory functions, with early symptoms often involving olfactory dysfunction – a decreased or distorted ability to smell. De novo Parkinson's disease refers to the initial onset of Parkinson's disease (PD) in an individual who has not previously been diagnosed with the condition and is not yet taking prescribed PD-specific medications.
According to the Parkinson’s Foundation, making an accurate diagnosis of Parkinson’s, particularly in its early stages, can be difficult. However, the earlier the disease is diagnosed, the earlier a person can receive interventions to slow its progression.
Researchers believe a combination of genetic and environmental factors cause Parkinson’s. In 2003, Heiko Braak, MD, hypothesized that an unknown pathogen in the gut, such as a bacteria, virus or other microorganism that causes disease, could be the cause of PD.
This was followed by a more extensive hypothesis, stating that PD starts in two places: the neurons of the nasal cavity and the neurons in the gut. This is now known as Braak’s hypothesis. In this theory, the pathogen enters the body via the nose and/or gets swallowed and reaches the gut. The pathogenic products thus come into contact with the olfactory (smell) and/or enteric (gut) neurons, triggering the aggregation of an abnormal protein called α-Synuclein. The aggregated α-Synuclein (called Lewy body) then spreads toward the central nervous system (namely the brain), and eventually arriving in and causing the degeneration of the dopaminergic neurons in the area of the brain called the substantia nigra.
This theory is supported by evidence that non-movement symptoms of PD, such as a loss of sense of smell, sleep disorders and constipation, may appear several years ahead of movement symptoms. For this reason, researchers focus on these non-motor symptoms to detect PD as early as possible in order to slow the progression of the disease through timely interventions.
Given the importance of detecting these early biomarkers, this paper proposes the novel use of chemosensory electroencephalogram (EEG), a technique used to study brain activity in response to chemical stimuli, specifically those perceived by the sense of smell or taste as it captures the brain’s responses to olfactory stimuli, one of the primary sensory modalities affected by the disease. Unlike expensive clinical tests such as neuroimaging, EEGs are non-invasive, cost-effective, and can detect subtle neurological changes in early PD.
A diagram of EEG signals.
The proposed method employs an Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) to decompose EEG signals into intrinsic mode functions (IMFs), which are different oscillatory modes, each with its own frequency and spatial distribution.
Entropic features – signs of disorder – including approximate entropy (AEn), sample entropy (SEn), and Rényi permutation entropy (RpEn), are extracted from these IMFs to identify distinguishing characteristics and compare them to healthy control subjects. Looking at the degree of entropy can be useful because research by C. Guducu et al (2015) shows that healthy individuals show significant changes in entropy during olfactory stimulation, unlike PD patients.
This study included 20 newly diagnosed (de novo) unmedicated patients and 12 healthy controls. Each person was exposed to phenethyl alcohol, a colorless liquid with a pleasant floral odor, often described as rose-like. EEG signals were recorded with a sampling rate of 1 kHz using a 64-channel device. Data were recorded from standard electrode locations focusing on central and lateral regions for olfactory processing.
Improved Complete Ensemble Empirical Mode De-composition with Adaptive Noise (ICEEMDAN), a third generation EMD, ensures better mode separation and stability by adaptively refining the decomposition process, making it highly effective for analyzing non-stationary signals such as EEG. Its ability to control noise contamination and reduce residual interference significantly improves the accuracy of signal decomposition compared to the original EMD, where different frequency components were often incorrectly combined.
EEG signal samples from (a) PD and (b) Healthy subjects
After extracting the IMFs, three of the seven IMFs were fed as input into various classifiers to evaluate their performance. These classifiers included: Random Forest, Decision Tree, Ensemble Subspace k-Nearest Neighbors, Ensemble Bagged Trees, XGBoost, Least Squares Support Vector Machine.
A comprehensive evaluation in which a number of variations were studied revealed that combining Rényi permutation entropy (RpEn) features with the Least Squares Support Vector Machine classifier achieves optimal performance with an accuracy of 96.47%, a precision of 96.14%, and a kappa score of 0.95. The paper’s authors note that the classifier’s performance is dependent on high-quality, consistent data. Further, they state that study has limitations, including a small dataset that restricts generalizability and inter-subject variability in olfactory responses, affecting classification. Additionally, the lack of multimodal integration, such as fMRI (functional Magnetic Resonance Imaging) or fNIRS (functional Near-Infrared Spectroscopy), limits comprehensive biomarker assessment. The authors suggest that expanding datasets and incorporating multimodal approaches can enhance diagnostic accuracy of PD.
Interested in learning more about EEG Signals? The IEEE Xplore Digital Library offers over 24,000 publications on EEG Signals.
Interested in acquiring full-text access to this collection for your entire organization? Request a free demo and trial subscription for your organization.
