´Blind´ data calibration of intravascular ultrasound data for automated tissue characterization

Spectral analysis of intravascular ultrasound (IVUS) backscatter has been shown to provide accurate tissue characterization in combination with statistical classification schemes. Such schemes employ multiple spectral parameters at the same time for tissue characterization. Before calculating the parameters, the tissue spectra is normalized with a spectra calculated with data from an ultrasound reflector. While this practice is feasible and convenient ex vivo, it is not optimal for the in vivo catheterization laboratory environment. Also, the additional procedure of gathering ultrasound reflector data for each IVUS catheter enforces a post-processing approach to IVUS tissue characterization instead of the desired real-time results. Blind deconvolution (BD) is an iterative algorithm that deconvolves the catheter transfer function (CTF) from the backscatter, thus enabling automated data normalization. BD is implemented with a recursive algorithm via an iterated window maximization method. In brief, if the common convolution product was to be modified as time variant, it could be written such that the pulse shape depends on ´n´ and a fixed ´K´ where the CTF or h(k,n) := 0 for k<0 and k>K and n=1,2,...N, where h(.,n) is the pulse shape due to a reflector at position n. Here, noise is assumed to be zero mean Gaussian white noise, denoted as measurement noise, that is not accounted for in this model. Finally, in order to obtain a computationally efficient solution, the pulse is assumed to be invariant on small intervals. Data were collected ex-vivo from 61 coronary arteries, with 30 MHz IVUS in saline at physiologic pressure. Regions of interest (ROI), selected from histology, comprised 115 fibrous tissue (FT), 63 fibro-fatty (FF), 88 necrotic-core (NC) and 56 dense calcium (DC) regions. ROI spectra were normalized after BD and compared to those previously reported with inverse filtering. Statistical classification schemes were computed with spectral parameters and were used to assess the accuracy of BD. Predictive accuracies of these classification schemes were comparable to the previously reported ultrasound reflector technique and can be further optimized, improving data normalization and decreasing computation time.