Fig. 2.

Tumor appearance, size, and detection by manual and automated methods. (A-I) Photomicrographs of tumor showing brown EpCAM staining (tumor; A-C), pimonidazole staining (hypoxia; D-G) or collagen IV staining (vessels; G-I), each with blue cresyl violet counterstain at week 3 (A, D, and G), week 4 after sham radiation therapy (RT; B, E, and H), and week 4 after 25 Gy radiation therapy (C, F, and I). Photomicrographs are from adjacent 20 μm sections from the same animals. Scale bars are 1 mm for widefield panels and 100 μm for magnified insets. (J) Histologic tumor volume in each animal. (K) Pre- and post-radiation therapy tumor volumes, as assessed by independent observation of post-Gd T₁-weighted magnetic resonance imaging (MRI) data sets. (L) Tumor volume detected by each method relative to the mean histologic size within each of the 3 groups of animals; mean ± 95% confidence intervals (CI). Mean and 95% CI for histologic detection are solid and dashed horizontal lines, respectively; n = 12 in each group. (M) Jaccard indices between tumor maps produced by each of the 4 methods and histologic gold-standard tumor maps; n = 12 in each group. (N) Sensitivity and (O) specificity of each method for detecting voxels identified as tumor histologically; data are given as mean + 95% CI; n = 12 in each group. *P < .05, **P < .01, ***P < .001 relative to human observer; †P < .05, †††P < .001 relative to multimodal MRI, analysis of variance followed by Sidak multiple comparison test.