S. Lahmiri, C. Tadj, C. Gargourb, S. Bekiros
The
analysis of infant cry signals is becoming an attractive field of
research in biomedical physics and engineering for better understanding
of the pathologies and appropriate medial diagnosis. The main purpose
of the current study is to characterize infant normal and pathological
cry signals by studying their respective oscillations by means of
approximate entropy and correlation dimension estimated from their
respective cepstrums. We analyzed two different sets. The first one is
composed of 2638 expiration cry signals and the second set is composed
of 1860 inspiration cry signals, both sets equally weighted. After
estimating approximate entropy and correlation dimensions from
cepstrums, three standard statistical tests are applied to them
including the Student t-test, F-test, and two-sample Kolmogorov-Smirnov
test. All statistical tests are performed at 5% statistical
significance level. The empirical results follow. First, approximate
entropy and correlation dimension measures exhibit different
statistical characteristics across healthy and unhealthy infant cries
from both expiration and inspiration sets. Second, the level of
approximate entropy in cepstrums of healthy infant cries is
statistically higher than that in cepstrums of unhealthy infant cries.
Third, the level of correlation dimension in cepstrums of healthy
infant cries is statistically higher than that in cepstrums of
unhealthy infant cries. In other words, cepstrums of healthy infant
cries show lower randomness and disorder compared to cepstrums of
unhealthy infant cries. It is concluded that cepstrum-based approximate
entropy and correlation dimension discriminate healthy from
pathological infant cry signals and can be employed as effective
biomarkers for biomedical diagnosis of cry records in clinical milieu.
Keywords: Infant cry
signal; Cepstrum; Complexity; Approximate entropy; Correlation
dimension; Statistical Tests.