Brushlet segmentation for automatic detection of lumen borders in IVUS images: A comparison study

Due to high scattering effects inside lumen, detection of luminal borders in intravascular ultrasound (IVUS) images becomes challenging when high frequency transducers are employed. In this paper, we further study previously developed three-dimensional (3D) multiscale overcomplete brushlet-driven harmonic analysis, motivated by what experts visually do, to trace the lumen borders by exploiting spatial frame incoherence within blood speckle patterns. Two-dimensional (2D) brushlet coefficient clustering was designed to isolate blood pool and estimate the lumen borders with the surface function active (SFA) framework. We evaluated our proposed algorithm on phantom with flowing fluid and 1081 clinical IVUS images acquired from six patients with single-element 40 MHz and 45 MHz transducers. We quantified and compared the results with a threshold-based algorithm and a 2D shape-driven technique driven by non-parametric probabilistic energy functions. We highlight the advantages of each approach and discuss the robustness of proposed algorithm.