A visual analytics approach to diagnosis of breast DCE-MRI data

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) of the breast is the most sensitive image modality for the detection of invasive breast cancer. To increase the moderate specificity of DCE-MRI, and therefore, the distinction of benign and malignant tumors, the tumor's heterogeneity and the tumor's enhancement kinetics have to be evaluated. In clinical practice, the tumor's enhancement kinetics are analyzed via time-intensity curves after manual placement of regions of interest (ROI). A substantial limitation of the ROI analysis is the inter-observer variability as well as the potential distortion of the ROI's average curve, e.g. if the ROI covers benign and malignant tumor tissue. We present a visual analytics approach for breast tumors in DCE-MRI data that comprises a voxel-wise glyph-based overview and and a region-based analysis. The regions are extracted via region merging and each region contains voxels with similar perfusion characteristics. As a result, we avoid the inter-observer variability and reduce distortion due to averaging over differently perfused tissue. A comparative study of 20 datasets was carried out to test our approach and an adapted time-intensity curve classification method. Moreover, the influence of similarity measurements and a potential region-based exploration are discussed. In conclusion, the presented features as similarity criteria yield the best results regarding the finer classification of the early contrast agent accumulation and the region's enhancement kinetics in the intermediate and late postcontrast phase since spatial information is included and the merging of regions with different perfusion characteristics is impeded. (C) 2010 Elsevier Ltd. All rights reserved.