• 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • br was incorporated in the training folds there was


    was incorporated in the training folds, there was no sin-gle multi-parameter set. However, as depicted in Figure 5a, the selection of the parameters clearly favored a specific order of combination. Ultimately, only the GMM and SVM with combinations of respectively three and eight parameters were included in the analysis. Figure 5b shows the distribution of the GMM scores for all pathologies.
    In the past decade, CUDI has been shown to be a promising tool for the characterization of prostatic tissue based on 2-D DCE-US recordings (Kuenen et al. 2011; Kuenen M et al. 2013; Mischi et al. 2012; van Sloun et al. 2017b). 3-D analysis does not only mitigate the risk of missing PCa lesions outside of the conventional 2-D field of view but the clinical work flow is also substantially sped up because of the need for only one FLAG tag Peptide injection. 
    Prompted by the introduction 3-D DCE-US imaging, 3-D CUDI similarity analysis has recently been developed and validated against the presence of PCa in SBx cores (Schalk et al. 2018). Moreover, a novel method for 3-D convective-dispersion modelling was developed (Wildeboer et al. 2018). This work elaborates on the expansion of all CUDI algorithms into 3-D. Moreover, we assessed the correspon-dence of the 3-D CUDI parameters as well as their ability to discriminate sPCa from benign prostatic tissue.
    Interestingly, most 3-D parameters from different analyses had only a low to moderate correlation on a voxel-to-voxel value basis. There are however some distinctive resemblances between m and WIT, as well as between r, % and MI, which follow from sharing a common underlying model and physical principles. Convective dis-persion and system-identified dispersion seem remarkably uncorrelated. As hypothesized by Wildeboer et al. (2018), this might be the consequence of the former reflecting spa-tial dispersion, whereas the latter primarily captured
    Fig. 4. Distributions of region-mean parameter values for benign (B), benign prostatic hyperplasia (BPH), prostatitis (P), insignificant (i.e., Gleason 3 + 3) PCa and significant (i.e., Gleason 3 + 4) PCa. The gray dots depict data points, the boxplots reflect their distributions and the lines connect distributions that are significantly (*) or highly significantly (**) different. B = benign; BPH = benign prostatic hyperplasia; P = prostatitis; PCa = prostate cancer; iPCa = insignificant prostate cancer; sPCa = significant prostate cancer; TIC = time-intensity curve.
    temporal dispersion of the contrast bolus. We speculated that the moderate correspondence between v and vCD might be caused by spatial convection being partly captured by both vCD and DCD.
    PCa-induced angiogenesis has been considered a key discriminating factor for PCa imaging (Padhani et al. 2005) and, indeed for CUDI, to distinguish cancer from healthy tissue. However, benign prostatic diseases, such as 
    BPH and prostatitis, are also associated to some extent with angiogenesis (Sandhu 2008; Shih et al. 2003). Although BPH often exhibits an increase in microvascular density (Deering et al. 1995), also reflected by BPH hav-ing a significantly higher a in Figure 4, other parameters do not generally show suspicious values. In termof molec-ular angiogenic activity, BPH + P has been observed to have significantly elevated levels of vascular endothelial
    3-D CEUS for Prediction of Prostate Cancer R. R. WILDEBOER et al. 9
    Table 1. 3-D contrast-enhanced ultrasound parameters and their full-set single-parametric performance for both benign tis-sue versus PCa and benign tissue versus sPCa
    Symbol Parameter Name Unit
    PCa sPCa
    Model-Fit Analysis s 1
    k Dispersion Parameter
    m Mean Transit Time s
    a Area Under The TIC a.u.
    Similarity Analysis
    r Spatiotemporal Correlation