Acquisition, Analyses and Classification of Brainstem Auditory Evoked Responses

By H.C. Nagaraj

Abstract

Human brainstem auditory evoked responses (BAERs) are sensory evoked potentials that can be recorded within a few milliseconds following a transient click stimulus. This response is picked up from the scalp and represents the involvement of auditory and central pathways. It is a series of neuroelectric potentials recorded from electrodes paced on the scalp with response latencies within 10 ms of the onset of the stimulus. BEARs provide a sensitive tool for assessment of brainstem auditory tracts and diagnostic information in the differentiation of metabolic from structural causes of brainstem dysfunction. The limitation of most measurement systems is that the diagnosis of abnormal pathology is an informal process. Therefore, there is a need to perform the BAER measurement study on patients belonging to different classes to arrive at statistically significant results and also to categorize them clearly into those of normal and various pathological cases.

The BAER signals are recorded on normal subjects and patients with clinical complaints of vertigo, deafness, acoustic neuroma and multiple sclerosis. These signals are analyzed in time and frequency domains to evolve some indices to clearly classify them into normals and various patient groups. The effects of nonpathologic factors such as the technical and subjective on the features of the BAERs have also been studied. The power spectral estimation of all the recorded BAER signals using the Fast Fourier Transform (FFT) algorithm has been carried out. The BAER spectrum of normal subjects contains three main frequency components, i.e. low-, mid- and high- frequency components around 100, 500 and 1000 Hz, respectively, which is not so in the case of diseased subjects. The spectral parameters, namely, the mean power frequency (MPF), median frequency (MF), the ratios of the integrated power at dominant frequencies to that of the total power in spectrum and change in spectral power (CP) between these dominant frequency components are estimated and used to classify BAER signals clearly into those of normal subjects and the above mentioned pathological cases. A new index CP derived from BAER signals and its range for the different subject classes appear to be the most dominant and promising parameter in the classification of the BARE signals. The student's t-test and regression analysis have been performed on normal and patient data to determine if there is a statistically significant difference between them.

An auto regressive (AR) model has been established for the recorded BAER signals with an optimum model order. The AR coefficients are mapped onto the feature plane representation to observe the relationship between these coefficients. This feature plane representation helps in easy distinction of the subjects of different abnormal states. The poles are extracted from the system function of the AR model and plotted over a unit circle in the Z-plane. The poles of the patients scatter widely and move closer to the unit circle tending towards instability, whereas those of the normal cluster near the origin of the circle. The pole locations characterize the severity of the disease. The power spectrum is computed from these model coefficients for the normal and diseased cases, resulting in a smoother power spectrum compared with that obtained using FF algorithm. The AR spectrum demonstrates its ability in characterizing the dominant resonances in the spectrum clearly showing the concentration of powers at certain frequency components. These techniques are not only used to classify the various pathological classes but also aids in assessing the severity of the disease. Finally, this work will help the clinicians in quick, reliable, better diagnosis and in therapy planning.