br The aim of this work was
The aim of this work was to evaluate and compare 16 individual 3-D DCE-US/CUDI model-based parame-ters for the detection of PCa, as well as to examine the technical feasibility of combining their information to more accurately predict individual SBx outcomes. To this end, we describe the implementation and validation of both a single-parametric and a multi-parametric approach in the remaining sections of this paper.
MATERIALS AND METHODS
Clinical data acquisition
The 3-D DCE-US recordings were carried out at the Second Affiliated Hospital of Zheijang University (Hang-zhou, China). The study was approved by the Institutional Review Board. Participants were scheduled for standard 12-core SBx and signed an informed consent. Before biopsy, 2-min contrast imaging was performed using a RIC5-9 probe connected to a LOGIQ E9 ultrasonic scanner (GE HealthCare, Wauwasota, WI, USA). To this end, a 2.4-mL SonoVue (Bracco, Milan, Italy) contrast-agent Cyclosporin H was intravenously administered. The volumetric cine-loops were acquired with a ’’low’’ image quality set-ting to maximize the frame rate, and UCA disruption was avoided by limiting the output power (AO%) to 10. The
data were then projected to Cartesian coordinates with a voxel size of 0.25 mm and the frame rate was maximized by choosing the field-of-view as narrow as possible, overall being 0.27 Hz on average.
A total of 58 3-D DCE-US recordings were per-formed, but 15 acquisitions were excluded based on pro-tocol violations (seven), technical complications (six) or an inconclusive biopsy report (two). In the remaining set, a total of seven biopsy cores could not be scored (in three patients) and were left out of the analysis. All other biopsy cores were histopathologically examined (Mon-tironi et al. 2005) and marked as benign (B), BPH, pros-tatitis (P), Gleason score 3 + 3 insignificant prostate cancer (iPCa) or Gleason score 3 + 4 significant pros-tate cancer (sPCa). Cores that contained both tumor and BPH were classified as malignant, whereas cores that contained BPH as well as P were classified as P.
On a fundamental level, CUDI is based on the notion that the behavior of blood in the prostatic (micro)vascula-ture can be modelled by anisotropically dispersive, multi-trajectory fluid flows in porous media. As a consequence, the kinetics of the UCA concentration (referred to as C) obey the 3-D convection-dispersion equation (Ewing and Wang 2001; LaBolle et al. 1998; Whitaker 1967)
where we assume the convective 3 £ 3 dispersion tensor (D) and the 3 £ 1 velocity vector (v) to be locally constant.
There are several ways to exploit this theoretical frame-work to quantify the local convective-dispersive behavior in tissue, but all CUDI algorithms essentially rely on the voxel-specific contrast-intensity evolution over time (i.e., time-intensity curve [TIC]) to describe the UCA concentration at each voxel as the bolus passes. In a model-fit approach, each individual TIC is fitted by a local-density random walk (LDRW) model to quantify the local convective dispersion (Kuenen et al. 2011). Alternatively, the similarity in TIC shape between a voxel and its neighbors can serve as charac-terization of the dispersion in the vessel architecture connect-ing those voxels (Kuenen MPJ et al. 2013; Mischi et al. 2012; Schalk et al. 2017). In addition, a system identification approach can be employed to quantify this transition in terms of apparent velocity and dispersion in the underlying convec-tive-dispersion system (van Sloun et al. 2017b). These approaches will be discussed more elaborately in the remain-ing sections of the materials and methods.
Because of the low frame rate of volumetric acquisi-tions, it is more difficult to suppress noise before analysis and to mitigate the effect of UCA recirculation. Nevertheless, it was shown that volume rates of approximately 0.25 Hz