The key role of angiogenesis in cancer growth has motivated the development of several imaging strategies for detection of neo-angiogenic processes. Recently, analysis of the dispersion kinetics of ultrasound contrast agents (UCAs) has been proposed as a promising approach for prostate cancer (PCa) localization. Determined by multipath trajectories through the microvasculature, dispersion enables characterization of the microvascular architecture and, therefore, localization of neo-angiogenic processes related to cancer growth [1].
Spatiotemporal analysis of indicator dilution curves (IDCs) measured at each pixel by dynamic contrast-enhanced ultrasound imaging has been proposed to assess the local dispersion kinetics of UCAs [2]. In particular, the similarity between IDCs at neighbor pixels is analytically related with the dispersion coefficient according to the advection-dispersion equation [3]. However, only linear similarity measures, such as temporal correlation or spectral coherence, have been used [2,4].
In this work, we propose a general similarity measure based on information theory. After an intravenous injection of a 2.4-mL SonoVue® (Bracco, Milan, Italy) bolus, its passage through the prostate is imaged using an iU22 ultrasound scanner (Philips Healthcare, Bothell, WA). Local dispersion is then estimated by mutual information analysis of IDCs from neighbor pixels, defined by a kernel whose size is designed accounting for the imaging-system resolution and aiming at an optimal balance between diagnostic resolution and reliability. A preliminary validation was performed with 15 patients referred for radical prostatectomy at the AMC University hospital of Amsterdam (NL), by comparison with the corresponding histology results on a pixel basis.
Pixel classification using the presented method resulted in a sensitivity and specificity equal to 81% and 87%, respectively. The receiver operating characteristic (ROC) curve area was 0.92. These results outperformed those obtained by linear similarity measures, as well as by any other perfusion measure.
To conclude, contrast ultrasound dispersion imaging by mutual information shows promising results for PCa localization and motivates towards further validation with a larger dataset.
References
[1] Kuenen et al, IEEE TMI, 2011.
[2] Mischi et al, IEEE TUFFC, 2012.
[3] Kuenen et al, UMB, 2013.
[4] Kuenen et al, IEEE TUFFC, 2013.
