Page 38 - 陽明交大電機學院院刊 2025.02

IEEE Transactions on Neural Systems and Rehabilitation
Engineering

Decoding Human Somatosenso-ry Sensitivity Through
Resting EEG and Behavioral Analysis: A Multimodal
Fusion Approach

2024/09,

電控工程研究所 柯立偉 教授（通訊作者）

types of power spectral density (PSD), BB-PSD
This study proposed a fusion approach to
and FB-PSD, and two types of inter-electrode
decode human somatosensory sensitivity by
connectivity (IEC), BBC and FBC, derived from
combining multimodal features (quantitative
resting-state EEG were subjected to feature
sensory testing and neurophysiology). Sixty
selection with a proposed prior-compared
healthy participants were exposed to cold, heat,
minimum-redundancy maximum-relevance
mechanical punctate, and pressure stimuli, and
(PCMRMR) protocol. Using the SVM, kNN, RF,
resting-state EEG data were collected. The QST
and GC models, the best accuracy of 87% was
scores of the participants were clustered using the
obtained when classifying participants into HS,
unsupervised k-means algorithm into four
NS, TS, or MS groups. It indicated that quantified
subgroups: generally hypersensitive (HS),
multi-parameter somatosensory sensitivity could
generally non-sensitive (NS), predominantly
be achieved with acceptable accuracy, leading to
thermally sensitive (TS), and predominantly
considerable possibilities for using objective pain
mechanically sensitive (MS). Furthermore, two
perception evaluation in clinical practice.

37