Research

Adaptive signal decomposition

Univariate adaptive decomposition techniques like Empirical Mode Decomposition (EMD), Empirical Wavelet Transform (EWT), and Iterative Filtering are widely used across various research domains. However, their application to multichannel signal analysis is limited by challenges such as mode misalignment, loss of mutual information, and unequal oscillatory modes across channels. A significant drawback, mode mixing in univariate EMD, has led to the development of multivariate extensions to address these limitations and improve signal decomposition for multichannel data.

Univariate Iterative Filtering has been extended to multivariate signals by employing a unified moving average filter for all channels. This extension offers a substantial reduction in computational time compared to existing multivariate methods like multivariate empirical mode decomposition (MEMD) and multivariate variational mode decomposition (MVMD) while maintaining effective signal decomposition.

Neurological Disease Diagnosis

A central contribution of my doctoral research is the application of Multivariate Iterative Filtering (MIF) for early and reliable detection of neurological disorders such as schizophrenia and Parkinson’s disease.

For schizophrenia, I developed an adaptive multichannel EEG rhythm-separation technique, enabling the extraction of highly discriminative features that achieved superior classification accuracy.
For Parkinson’s disease, I proposed a novel phase-space–based feature derived from MIF-decomposed signals, effectively capturing complex spatiotemporal patterns.

Both approaches outperformed state-of-the-art methods in terms of sensitivity, specificity, and robustness. These frameworks pave the way for computationally efficient, non-invasive diagnostic tools that can directly support clinical decision-making.

Brain-Computer Interface

Another major contribution of my work is the development of MIF-driven pipelines for enhancing Brain-Computer Interface (BCI) systems, aimed at assisting communication and motor control in individuals with severe impairments.

I designed a hybrid MIF–Common Spatial Patterns (MIF-CSP) framework for motor imagery BCIs, significantly boosting the accuracy of classifying imagined motor tasks.

I further proposed an MIF–Canonical Correlation Analysis (MIF-CCA) method for steady-state visual evoked potential (SSVEP)-based BCIs, validated in mobile scenarios where subjects performed tasks while walking at different speeds.

These methods demonstrated strong resilience against artifacts and environmental noise, underlining their potential for deployment in portable, real-world neurotechnologies that restore autonomy and improve quality of life.

Page updated

Google Sites

Report abuse