Optimization of attenuation estimation in reflection for in vivo human dermis characterization at 20 MHz.

In vivo skin attenuation estimators must be applicable to backscattered radio frequency signals obtained in a pulse-echo configuration. This work compares three such estimators: short-time Fourier multinarrowband (MNB), short-time Fourier centroid shift (FC), and autoregressive centroid shift (ARC). All provide estimations of the attenuation slope (beta, dB x cm(-1) x MHz(-1)); MNB also provides an independent estimation of the mean attenuation level (IA, dB x cm(-1)). Practical approaches are proposed for data windowing, spectral variance characterization, and bandwidth selection. Then, based on simulated data, FC and ARC were selected as the best (compromise between bias and variance) attenuation slope estimators. The FC, ARC, and MNB were applied to in vivo human skin data acquired at 20 MHz to estimate betaFC, betaARC, and IA(MNB), respectively (without diffraction correction, between 11 and 27 MHz). Lateral heterogeneity had less effect and day-to-day reproducibility was smaller for IA than for beta. The IA and betaARC were dependent on pressure applied to skin during acquisition and IA on room and skin-surface temperatures. Negative values of IA imply that IA and beta may be influenced not only by skin's attenuation but also by structural heterogeneity across dermal depth. Even so, IA was correlated to subject age and IA, betaFC, and betaARC were dependent on subject gender. Thus, in vivo attenuation measurements reveal interesting variations with subject age and gender and thus appeared promising to detect skin structure modifications.