Tumor heterogeneity complicates the quantification of a therapeutic response by MRI. To address this issue, a novel approach has been developed that combines MR diffusion imaging with multispectral (MS) analysis to quantify tumor tissue populations. K-means (KM) clustering of the apparent diffusion coefficient (ADC), T2, and proton density (M0) was employed to estimate the volumes of viable tumor tissue, necrosis, and neighboring subcutaneous adipose tissue in a human colorectal tumor xenograft mouse model. In a second set of experiments, the temporal evolution of the MS tissue classes in response to therapeutic intervention Apo2L/TRAIL and CPT-11 was observed. The multiple parameters played complementary roles in identifying the various tissues. The ADC was the dominant parameter for identifying regions of necrosis, whereas T2 identified two necrotic subpopulations, and M0 contributed to the differentiation of viable tumor from subcutaneous adipose tissue. MS viable tumor estimates (mean volume = 275 +/- 147 mm(3)) were highly correlated (r = 0.81, P < 0.01) with histological estimates (117 +/- 51 mm(3)). In the treatment study, MS viable tumor volume (at day 10) was 77 +/- 67 mm(3) for the Apo2L/TRAIL+CPT-11 group, and was significantly reduced relative to the control group (292 +/- 127 mm(3), P < 0.01). This method shows promise as a means of detecting an early therapeutic response in vivo.
Copyright 2004 Wiley-Liss, Inc.