Towards Extra-Luminal Blood Detection from Intravascular Ultrasound Radio Frequency Data

Abstract

Abstract. Recent evidence has suggested that the presence and proliferation of vasa vasorum (VV) in the plaque is correlated to an increase in plaque inflammation and destabilization, leading to acute coronary events (e.g., heart attacks). Therefore, the detection and quantification of VV in plaque (i.e., extra luminal blood perfusion) is an important problem since it may enable the development of an index of plaque vulnerability. In this paper, we explore the feasibility of a method that employs a physics-based model of the scattered intravascular ultrasound (IVUS) radio frequency signal for the detection of blood. We evaluate our method using synthetic data and validate it using six 40 MHz pullback sequences acquired with three different IVUS systems from different arteries of rabbits and swines. Our experimental results are very promising and indicate the feasibility of our method for the computation of a feature that leads to automatic extra-luminal blood detection which may be an indication of plaque inflammation.

Probabilistic Segmentation of the Lumen from Intravascular Ultrasound Radio Frequency Data

“…For each sequence, we selected five groups of seven consecutive frames from different periods of time for a total of 35 frames per sequence. The parameters used for the computation of the DBC are σ = 5.3e −8 for the width of the envelope of the impulse function, μ = 0.08276 dB/mm corresponding to the attenuation coefficient of blood, c = 1, 540 × 10 3 mm/s corresponding to the speed of sound in biological tissue, P = 0.05 mm for the size of partition, D = 400 scatterers/mm −2 for the scatterer density, and δ = 3 for the cardinality of neighbors according to [17]. The number of coefficients used for the contour parameterization was N k = 5.…”

“…For the computation of the a priori terms, we employ features extracted from the IVUS RF signal using our blood detection method proposed in [17]. In that work, the structures present in the vessel are modeled as a distribution of random positioned scatterers for which the acoustic power scattered in the direction of the ultrasound transducer is defined by the differential backscattering cross section (DBC) of the scatterer.…”

“…In summary, our method consists of six steps: (i) divide the RF signal corresponding to each A-line of a frame into N P non-overlapping partitions; (ii) compute the DBC values for all the partitions using the method described in [17]; (iii) construct a feature vector v p ∈ R 2 for each partition p by concatenating the DBC computed for that partition and its radial distance from the transducer; (iv) classify each partition as blood or non-blood according to the prediction of a classifier; (v) use the confidence of the classification results as the a-priori term in the Bayesian optimization function; and (vi) minimize the cost function that guides the position and shape of the lumen contour using a line-search method.…”

“…RF-based approaches include methods for the characterization of different regions of interest such as plaque or blood (e.g., [16,17]). …”