Analysis of Clinical Percussion Signals Using Matching Pursuit

Abstract

Abstract:Clinical percussion is a method of eliciting sounds from the body by tapping with either a percussion hammer or fingers to determine the area under the perused is air filled, fluid filled, or solid, and is used in clinical examinations to assess the condition of the thorax or abdomen. Successful diagnosis today is still highly subjective and dependent's on physician skill, experience and require quite surrounding areas. An automated system capable of delivering standardized percussion analysis would remove these limitations on the technique and allow for its usage by those without such specialized training and years of necessary experience. For this to be possible, efficient and informative signal processing algorithms must be employed. In this investigation, clinical percussions from healthy volunteers taken by trained medical professionals were analysed via the matching pursuit (MP) algorithm. Various types of possible dictionaries are discussed comparing their efficiency and convergence behaviour. Noise filtering methods are discussed and a noise reduction method based on MP analysis results is presented.MP is also compared to other methods for representing clinical percussions with regards to informativeness and efficiency. MP is ( log ) which is more efficient than current methods which have a complexity of at least ( 3 ).

Comparison of Modelled Pursuits with Esprit and the Matrix Pencil Method in the Modelling of Medical Percussion Signals

“…There are methods that use a single orthogonal transform such as Fourier [2] or wavelet [8] to analyse the signal in one pass, but although these are cheap in terms of processing requirements, they typically perform poorly [13]. Matching Pursuits (MP) [4,14] is an iterative approach to signal decomposition that can use any dictionary (or dictionaries) of atoms. These dictionaries can be (and usually are) 'overcomplete' (i.e.…”

“…• Generate M test synthetic EDS components with parameters randomly selected from ranges that match the type of MPS as observed in [2,4]. φ = atan2(imag(amplitudes),real(amplitudes));…”

“…A variety of percussion signal capture and analysis techniques are described in the literature [e.g. 1,2,4]. These techniques model the captured medical percussion signals (which are the response of the body to the percussive input) as sums of exponentially decaying sinusoids (EDS), since this model is an intuitive and efficient way of describing the impulse responses (IRs) of vibrational systems.…”

“…In a digital system a single non-zero sample is an impulse that contains equal energy at all frequencies up to the Nyquist limit. Medical percussion involves introducing energy into the bodily region of interest by tapping with the fingers [4]. A single tap is an impulse-like signal (although it is certainly not perfectly impulsive -it will likely be much longer in duration than a single nonzero sample and therefore will have less energy at high frequencies due to the integrating effect of that longer duration), and so is likely to excite all vibrational modes of the object, or objects, being investigated.…”

“…In [4] the author states that 'a small number of low frequency atoms can be sufficient for classification of MPS'. In cases where it is known a priori that the number of EDS in the original signal is small, a modelling process can be used that only calculates the first few most significant atoms rather than the maximum possible by that modelling method.…”