Principal component analysis of breast DCE-MRI adjusted with a model-based method.

PURPOSE: To investigate a fast, objective, and standardized method for analyzing breast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) applying principal component analysis (PCA) adjusted with a model-based method. MATERIALS AND METHODS: 3D gradient-echo DCE breast images of 31 malignant and 38 benign lesions, recorded on a 1.5T scanner, were retrospectively analyzed by PCA and by the model-based three-timepoints (3TP) method. RESULTS: Intensity-scaled (IS) and enhancement-scaled (ES) datasets were reduced by PCA yielding a first IS-eigenvector that captured the signal variation between fat and fibroglandular tissue; two IS-eigenvectors and the two first ES-eigenvectors captured contrast-enhanced changes, whereas the remaining eigenvectors captured predominantly noise changes. Rotation of the two contrast-related eigenvectors led to a high congruence between the projection coefficients and the 3TP parameters. The ES-eigenvectors and the rotation angle were highly reproducible across malignant lesions, enabling calculation of a general rotated eigenvector base. Receiver operating characteristic (ROC) curve analysis of the projection coefficients of the two eigenvectors indicated high sensitivity of the first rotated eigenvector to detect lesions (area under the curve [AUC] > 0.97) and of the second rotated eigenvector to differentiate malignancy from benignancy (AUC = 0.87). CONCLUSION: PCA adjusted with a model-based method provided a fast and objective computer-aided diagnostic tool for breast DCE-MRI.

Scan-rescan variability in perfusion assessment of tumors in MRI using both model and data-derived arterial input functions.

PURPOSE: To evaluate the contribution to scan-rescan coefficient of variation (CV) of patient-specific arterial input function (AIF) measurement in dynamic contrast-enhanced MRI (DCE-MRI) data, and to determine whether any advantage or disadvantage to using a data-derived arterial input function is related to the anatomical location of the target lesion. MATERIALS AND METHODS: Two methods are presented for the calculation of perfusion parameters from DCE-MRI data using a two-compartment model. The first method makes use of a single-model AIF across all study data sets, while the second uses an automated process to derive an AIF specific to each data set. Both methods are applied to the analysis of a 25-subject scan-rescan study of patients with advanced solid tumors located in either the lungs or the liver. The parameters of interest in this study are the volume transfer constant between arterial plasma and extracellular extravascular space (Ktrans) and the blood-normalized initial area under the tumor enhancement curve over the first 90 seconds postinjection (IAUCBN90). RESULTS: The use of a data-derived AIF reduces the visit-to-visit CV in both parameters for liver lesions by approximately 70% while the improvement is less than 20% for lung lesions. CONCLUSION: The use of a data-derived AIF in the analysis of DCE-MRI data provides a substantial reduction in scan-rescan CV in the measurement of vascular parameters such as Ktrans and IAUCBN90. These results show a much larger advantage in the liver than in the lungs. However, this difference is largely driven by a small number of outliers, and may be spurious.