Feature extraction and state identification in biomedical signals using hierarchical fuzzy clustering

Area of Science:

• Biomedical signal processing
• Machine learning
• Pattern recognition

Background:

• Biomedical signal processing often involves state recognition and event prediction from complex data.
• Existing methods struggle with the non-stationary and discontinuous nature of biological signals like ECG and EEG.
• Accurate prediction of critical events, such as epileptic seizures or driver fatigue, remains a significant challenge.

Purpose of the Study:

• To develop an advanced clustering method for analyzing biomedical signals.
• To enable accurate state recognition and event prediction in biological systems.
• To address the non-stationary nature of biomedical data using fuzzy logic and hierarchical clustering.

Main Methods:

• Application of fuzzy clustering to group discontinuous temporal patterns in continuously sampled measurements.
• Development of a recursive algorithm for hierarchical fuzzy partitioning.
• Adaptive selection of the number of clusters to handle signal non-stationarity.
• Integration of hierarchical and fuzzy concepts for robust cluster validity.

Main Results:

• The proposed algorithm effectively groups related temporal patterns.
• Fuzzy clustering naturally handles transitions between biological states.
• The method demonstrates effectiveness across diverse datasets with varying characteristics.
• Successful application to state recognition in heart rate recovery and epileptic seizure forecasting from EEG.

Conclusions:

• The novel hierarchical fuzzy clustering algorithm provides a robust solution for biomedical signal analysis.
• The method enhances the ability to recognize states and predict events, including critical health events.
• This approach offers a feasible and effective tool for advancing biomedical signal processing applications.