Discrimination of Healthy and Post-Partum Subjects using Wavelet Filterbank and Auto-Regressive Modelling

Keywords: Wavelet, Autoregressive Modelling, Patient Discrimination, Pelvic Floor Muscle.

Rehabilitation therapies to treat female stress urinary incontinence focus on the reactivation of pelvic floor muscle (PFM) activity. An objective measure is essential to assess a subjet's imprvement in PFM capabilities and increase the success rate of the therapy. In order to provide such a measure, we propose a method for the discrimination of healthy subjects with strong PFM and post-partum subjects with weak PFM. Our method is based on a dyadic discrete wavelet decomposition of electromyograms (EMG) that projects slow-twitched and fast-twitched muscle activities onto different scales. We used a parametric autoregressve (AR) model for estimation of the frequency of each wavelet scale to overcome the poor frequency resolution of the dyadic decomposition. The feature used for discrimination was the frequency of the wavelet scale with the highest variance after interpolation with the nearest neighboring scales. Twenty-three healthy and 26 post-partum women with weak PFM who executed 4 maximum voluntary contractions (MVC) were retrospectively analysed. EMGs were recorded using a vaginal probe. The proposed method has a lower rate of false discrimination (4%) compared to the two classical methods based on mean (9%) and median (7%) frequency estimation from the power spectral density.