[home] [Personal Program] [Help]

---

Session: Cancer
Session starts: Friday 25 January, 10:30
Presentation starts: 11:00
Room: Lecture room 559

---

Rogier R Wildeboer (Eindhoven University of Technology)
Ruud JG van Sloun (Eindhoven University of Technology)
Pintong Huang (Second Affiliated Hospital of Zhejiang Hospital, China)
Hessel Wijkstra (Amsterdam Medical Centers, University of Amsterdam )
Massimo Mischi (Eindhoven University of Technology)

Abstract:
Introduction ♦ Today, non-targeted ≥10-core systematic biopsy (SBx) using transrectal ultrasound (TRUS) is the guideline-recommended clinical pathway for prostate cancer (PCa) diagnosis. However, this technique is associated with a risk of underdiagnosis and overtreatment as well as the occurrence of complications. Three-dimensional (3D) dynamic contrast-enhanced ultrasound (DCE-US) recordings exploit specific imaging of contrast agents to visualize and characterize (micro)vascularity. Quantification algorithms for DCE-US of the prostate have shown good potential for PCa localization in 2D. With the introduction of 3D DCE-US, a few have recently been expanded to three dimensions. Materials and Methods ♦ We present the 3D implementation of a large range of estimators comprising contrast ultrasound dispersion imaging (CUDI) and assess their performance to discriminate prostate regions with and without PCa. Furthermore, we utilize combinations of the extracted parameters to predict individual SBx-core biopsy pathology. To this end, we apply a discriminative and a generative machine-learning approach, that is, a support vector machine (SVM) and a Gaussian Mixture Model (GMM), to a dataset of 43 2-minute 3D DCE-US recordings acquired at the Second Affiliated Hospital of Zhejiang University in Hangzhou, China. The machine-learning algorithms were trained and tested in a leave-one-prostate-out fashion. Results ♦ Individually, the estimators show good correlation with the presence of prostate cancer in SBx-regions. Best performing for PCa and significant PCa (sPCa) are convective velocity (ROC curve area PCa = 0.71; sPCa = 0.80), wash-in time (PCa = 0.71; sPCa = 0.78) and mean transit time (PCa = 0.69; sPCa = 0.79). The best machine-learning approach is the GMM, showing a ROC curve area of 0.76 and 0.81 for PCa and sPCa, respectively. Discussion and Conclusions ♦ 3D quantification of DCE-US for the characterization of prostate tissue yields promising results. Moreover, we demonstrated that machine-learning approaches can improve the classification performance compared to individual DCE-US parameters. Although this study is limited by the relatively small number of patients, these results show potential for further development in ultrasound-based PCa localization.