Dynamic contrast-enhanced (DCE) imaging for tumor delineation in prostate cancer

Abstract

Dynamic contrast-enhanced (DCE) MR imaging is frequently used for the detection and localization of prostate tumors. After injection of a bolus of contrast agent into the blood circulation, the behavior of the contrast agent in the prostate can be measured by repetitive imaging of the prostate. Prostate tumors are characterized by abnormalities in the blood flow and vessel permeability, which are reflected in abnormal behavior of the contrast agent. In a radiation treatment setting, accurate quantification of DCE data is needed, to allow for robust tumor delineation, tumor characterization and longitudinal studies. For the quantification of DCE data with a tracer kinetic model, the measurement of an arterial input function (AIF) is required. However, measuring an AIF directly from the DCE-MRI magnitude signal can be challenging. To prevent large errors in an exam-specific AIF, the use of a population-averaged AIF has been proposed for analysis of DCE-MRI data of the prostate. We found that for longitudinal and multi-center studies of quantitative DCE-MRI, the use of a population-averaged AIF is advantageous when the measurement precision of an exam-specific AIF is considerably worse than 15%. As an alternative to the magnitude signal, the AIF can also be measured from the DCE-MRI phase signal (AIFPHASE). Although some phantom and simulation studies have been performed, validation of AIFPHASE measurements is lacking. We found in a group of 12 prostate cancer patients that robust quantification of Ktrans values from DCE-MRI exams in the cancerous prostate is feasible with the use of AIFPHASE. For reliable identification of small lesions within the prostate, it is crucial that they are consistently detected when the examination and analysis are repeated. The smallest detectable lesion size, however, depends on the spatial resolution. Therefore, we evaluated the relationship between image noise, voxel size, and voxel-wise repeatability of DCE-CT examinations for prostate cancer and found that there is a high voxel-wise repeatability of the DCE-CT imaging technique for kernel sizes as small as 0.1 cm3. A number of different tracer kinetic models (TKM) have been clinically used to quantify the microvascular properties of the prostate. We investigated which of three frequently used TKMs is the optimal model for quantification of DCE-CT data. The adiabatic approximation to the tissue homogeneity (AATH) model seems the optimal model for quantification of DCE-CT data of the prostate. Delineating a prostate tumor essentially comes down to a voxelwise decision whether a voxel contains tumor or not. However, the sensitivity and specificity of the DCE technique are not perfect and by definition no detailed spatial verification of imaging with pathology can be obtained from patients scheduled for radiotherapy. In the clinical practice of radiotherapy treatment planning this means that there will never be a ground truth when delineating a prostate tumor. We propose a method to incorporate the uncertainty that a voxel contains tumor into the tumor delineation process. In conclusion, a number of studies have been performed to facilitate well-founded decision making in the quantification process of DCE-MRI and DCE-CT exams for tumor delineation in prostate cancer.