Delineation and segmentation of cerebral tumors by mapping blood-brain barrier disruption with dynamic contrast-enhanced CT and tracer kinetics modeling - a feasibility study

Dynamic contrast-enhanced (DCE) imaging is a promising approach for in vivo assessment of tissue microcirculation. Twenty patients with clinical and routine computed tomography (CT) evidence of intracerebral neoplasm were examined with DCE-CT imaging. Using a distributed-parameter model for tracer kinetics modeling of DCE-CT data, voxel-level maps of cerebral blood flow (F), intravascular blood volume (v(i)) and intravascular mean transit time (t(1)) were generated. Permeability-surface area product (PS), extravascular extracellular blood volume (v(e)) and extraction ratio (E) maps were also calculated to reveal pathologic locations of tracer extravasation, which are indicative of disruptions in the blood-brain barrier (BBB). All maps were visually assessed for quality of tumor delineation and measurement of tumor extent by two radiologists. Kappa (K) coefficients and their 95% confidence intervals (CI) were calculated to determine the interobserver agreement for each DCE-CT map. There was a substantial agreement for the tumor delineation quality in the F, v(e) and t(1) maps. The agreement for the quality of the tumor delineation was excellent for the vi, PS and E maps. Concerning the measurement of tumor extent, excellent and nearly excellent agreement was achieved only for E and PS maps, respectively. According to these results, we performed a segmentation of the cerebral tumors on the base of the E maps. The interobserver agreement for the tumor extent quantification based on manual segmentation of tumor in the E maps vs. the computer-assisted segmentation was excellent (K=0.96, CI: 0.93-0.99). The interobserver agreement for the tumor extent quantification based on computer segmentation in the mean images and the E maps was substantial (K=0.52, CI: 0.42-0.59). This study illustrates the diagnostic usefulness of parametric maps associated with BBB disruption on a physiology-based approach and highlights the feasibility for automatic segmentation of cerebral tumors.