The effects of personalized tPBM based on Machine Learning-based QEEG phenotypes

1651 

Abstract Submission 

Juhee Ko¹, Ukeob Park², Boeun Choi³, Seung Wan Kang³

¹iMediSync Inc., Seoul, Seoul, ²iMediSync, Inc., Seoul, seoul, ³iMediSync, Seoul, Seoul

Juhee Ko  
iMediSync Inc.
Seoul, Seoul

Ukeob Park  
iMediSync, Inc.
Seoul, seoul

Boeun Choi  
iMediSync
Seoul, Seoul

Seung Wan Kang  
iMediSync
Seoul, Seoul

Over the past 40 years, numerous studies have clustered Quantitative Electroencephalogram(QEEG) into phenotypes for clinical and research purposes.[1,2,3,4] Particularly, phenotypes based on the alpha band are notable.[5,6] Furthermore, recent research has been actively conducted using QEEG and Transcranial photobiomodulation(tPBM) for the treatment and diagnosis of various brain diseases.[7,8] However, it remains unclear how the effects of tPBM manifest in terms of EEG, cognitive function, and emotions based on the power spectral density(PSD) pattern of the occipital lobe. In this study, clustering was performed using a machine learning algorithm(ML)(Gaussian Mixture Method(GMM)+Support Vector Machine(SVM)) based on the PSD phenotype of the occipital lobe. Subsequently, a suitable tPBM protocol was applied on a case-by-case basis according to the characteristics of each cluster, and we provide insights into the altered QEEG pattern and the effects on cognitive function and emotions after tPBM.

We utilized EEG data measured at 19 channels on the 10-20 system in the resting state from a total of 104 subjects. Both eyes-closed and eyes-open data were employed as the dataset for the clustering model. As QEEG features used in the model, we utilized spectrum power and power ratio from the occipital lobe. Additionally, the participants include both normal and abnormal data. Among these 104 data, 80% were used as the train set, and the remaining 20% were used as the test set. We applied the Clustering Method (GMM) to select meaningful features for forming clusters, and subsequently trained the model using those features previously chosen as features for the Classification Method (SVM).The features of each cluster classified by the clustering model were used as criteria for labeling, and subsequently, these labels were employed as criteria for evaluating the performance of the classification model. The labels used were based on the commonly used QEEG phenotype criteria[1]. Cluster 1 corresponds to Persistent eyes-open alpha in the occipital lobe, Cluster 2 represents Low-voltage fast alpha in the occipital lobe during eyes-closed states, Cluster 3 includes complex abnormal patterns, and Cluster 4 denotes normal patterns (presence of alpha in the occipital lobe during eyes-closed states, and more than a 50% decrease in alpha during eyes-open states).Using the trained model, we classified 58 subjects into four groups and applied personalized tPBM protocols for each group over 8 weeks (3 sessions per week, 10 minutes per session). Subsequently, we evaluated the personalized tPBM effects on individualized changes in EEG, cognitive functions (comprehensive memory, inference ability, executive function, processing speed), and emotions (depression, stress, trait anxiety, state anxiety) through EEG examinations and questionnaires.

The accuracy of the classification model on the training set exhibited an F1 score of 0.88, Precision of 0.88, and Recall of 0.88. On the test set, the model showed an F1 score of 0.86, Precision of 0.86, and Recall of 0.88. Furthermore, the effects of tPBM for each cluster based on the results of the trained model showed a significant increase (p-value < 0.01) in processing speed and executive function among cognitive function indicators. Additionally, among emotional indicators, a significant decrease (p-value < 0.05) was observed in depression, state anxiety, trait anxiety, and stress.

This study, through a machine learning-based clustering model using QEEG phenotypes as a foundation, identified meaningful clusters and demonstrated significant positive effects on cognitive function and emotions through the implementation of personalized tPBM based on these clusters. Such personalized tPBM can be more effectively utilized for the treatment of cognitive function and emotions. Further development of diverse QEEG phenotypes is expected to broaden its application as a biomarker for brain diseases.

Classification and Predictive Modeling ²

EEG/MEG Modeling and Analysis ¹

EEG

NIRS

Anxiety

Attention Deficit Disorder

Cognition

Computational Neuroscience

Data analysis

Data Organization

Electroencephaolography (EEG)

Machine Learning

Modeling

Treatment