Whole-body MRI combined with a semiautomated hierarchical multispectral image analysis technique was evaluated as a method for detecting viable tumor tissue in a murine model of metastatic breast cancer (4T1 cell line). Whole-body apparent diffusion coefficient, T(2), and proton density maps were acquired in this study. The viable tumor tissue segmentation included three-stage k-means clustering of the parametric maps, morphologic operations, application of a size threshold, and reader discrimination of the segmented objects. The segmentation results were validated by histologic evaluation, and the detection accuracy of the technique was evaluated at three size thresholds (15, 100, and 500 voxels). The accuracy was 88.9% for a 500-voxel size threshold, and the area under receiver operating characteristic curve was 0.84. The regions of segmented viable tumor tissue within the primary tumors were found mostly on the periphery of the tumors in agreement with the histologic findings. The presented technique was found capable of detecting metastases and segmenting the viable tumor from necrotic regions within tumors found in this model. It offers a noninvasive, whole-body, viable tumor tissue detection method for preclinical and potentially clinical applications such as tumor screening and evaluating therapeutic efficacy.
(c) 2009 Wiley-Liss, Inc.