Use of Sparse Representation of ECG Signals for Pain Detection in the Cold Pressor

Introduction: Currently there is no objective measure
for pain. Heart rate has been studied using variability
(HRV) and spectral analysis for the assessment of pain
(1), but no clinically applicable tool is available (2). Traditional signal analysis methods involve Fourier representation or Wavelets representation. Digital signal
processing (DSP) uses the idea of merging several different methods to create a so-called over-complete dictionary. These methods decompose a signal into few
meaningful components by searching for the sparsest
possible representation, given a wise choice of over-complete dictionary. Methods: After IRB approval, 15
healthy young adult volunteers (10M, 5F) were subject
to a cold pressor test. Each subject was evaluated by
continuous ECG sampled at 1000 Hz for 5 minutes prior
to the test as baseline. Pain was rated on the VAS until
the pain tolerance was reached. Data were analyzed using
an over complete dictionary consisting of Fourier and
Wavelet transforms and the Orthogonal Matching
Pursuit algorithm (3). For all subjects we evaluated
30-mixed coefficients, with an average ratio of 1:1.2
Wavelet to Fourier coefficients. Results: Wavelet coefficients in the sparse representation of R-R signals are
significantly correlated with the periods of acute pain
development (Vas 0 to Vas 9) and acute pain relief (P <
0.0001). During constant pain, the wavelets coefficients
were not apparent. Conclusion: Our findings support
the usefulness of an over complete dictionary analysis of
the ECG signal as an immediate and reliable measure of
acute pain state changes.