Automation of pattern recognition analysis of dynamic contrast‐enhanced MRI data to characterize intratumoral vascular heterogeneity
Using preclinical in vivo imaging modalities coregistered with pathology, we have shown previously that well‐vascularized (i.e., well‐perfused) tumor areas are characterized by rapid Gd‐DTPA uptake/washout, that hypoxic areas exhibit reduced vascular function associated with delayed Gd‐DTPA uptake/washout, and that necrotic areas exhibit slow or no CA uptake and no discernible washout over the experimental observation 12. We categorized these tumor microenvironments based on their representative DCE‐MRI signal‐versus‐time curves by pattern recognition (PR), using the Gaussian mixture model or constrained nonnegative matrix factorization (cNMF) 24. The semiautomatic PR approach required manual input of the number of patterns (NP) in the DCE‐MRI data. The variable (subjective) application of a fixed NP for all tumor slices may lead to overfitting (or underfitting) in tumors or tumor slices that are characterized by more (or fewer) patterns than predefined, thus disregarding intratumoral heterogeneity represented by disparate DCE‐MRI curves and physiological environments across tumor slices (Fig. 1).
The goal of this study was to optimize and automate DCE‐MRI data analysis using our previously described unsupervised PR approach 24 to accurately and fully automate the identification of vascularity‐driven intratumoral heterogeneity using cNMF. This model involves novel automatic approaches to determine the NP for each DCE‐MRI slice, to spatially map intratumoral heterogeneities, and to incorporate the computerized determination of the precontrast signal. A stepwise scheme of the analysis process is shown in Figure 2. All analysis steps were coded in MATLAB (MathWorks, Inc. Natick, Massachusetts, USA).