Breast tumor classification and visualization with machine-learning approaches

Publisher Summary This chapter focuses on image sequences obtained by using protocols that generate a time series of ns = 6 volumes with scanners equipped with dedicated surface coils enabling simultaneous imaging of both female breasts. The time interval between the single volumes varies between 90 sec and 110 sec depending on the particular protocol. The in-plane resolution of the volume is between 1.4 mm × 1.4 mm and 1.2 mm × 1.2 mm. An experienced radiologist evaluates the 4D volumes in clinical practice by visualizing the uptake and wash-out characteristics of the tissue. The usual approach is a voxel-wise computing of signal differences in a volume pair. In clinical patient management, there is only limited time for evaluating dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) data. Such an evaluation includes three visual tasks: detection of suspicious lesions, classification of signals and lesions, and determination of tissue change. A sophisticated approach, three time points (3TP), is based on the idea of simultaneously visualizing both the signal uptake and washout characteristics in one display using a pseudocoloring algorithm. This algorithm maps the temporal signal information associated with each voxel to a color hue and an intensity value. The intensity value is computed from the signal uptake and the color hue from the washout. The recent development of commercial digital MR imaging systems provides a significant improvement in image quality for traditional screening and also information for the analysis and detection of mammographic features by computer. Contrast agent uptake descriptors are often single measured values that can be accessed directly utilizing a properly defined mathematical prescription consisting of one or more formulas.