Several previous studies have discussed the problem of intratumor vascular heterogeneity in compartment modelling, a major outstanding issue for the characterization of complex phenotypes and therapeutic responses. Some methods have addressed the estimation of multi-compartment pharmacokinetics in the presence of varying partial-volume effects, relying on known regions of pure-volume pixels and number of compartments. The significant advantage of our strategy is its ability to detect and quantify intratumor vascular heterogeneity without any type of external information. The benefits of such a method include its wide applicability, sensitive detection of heterogeneity dynamics, and reliance on longitudinal data from one single subject. We have identified differential and heterogeneous changes in tissue-specific vascular pharmacokinetics in tumors during treatment that were undetected using standard analysis, including tumor islands of persistent enhancement that have escaped the effects of therapy. These results are particularly intriguing when considered together with recent imaging studies describing foci of resistant and more aggressive clones within a tumor. While it is not yet possible to assign causality, these in vivo results allowed us to propose new hypotheses regarding the complex relationships VE-821 between intratumor heterogeneity, clonal repopulation, cancer stem-cell, and therapeutic efficacy. In metastatic disease, recent studies have revealed the emergence of treatment-resistant subclones that were present at a minor frequency in the primary tumour. Thus, modeling cancer diagnosis and treatment in the future should involve characterization of subpopulations within the primary tumour, monitoring of clonal dynamics during treatment and eradication of treatmentemergent clones. To prospectively assess intratumor heterogeneity, profiling of multiregional tumour samples would be required. However, it is impractical and potentially risky to take multiple ‘random’ biopsies in every patient, owing to both sampling bias and the inability to resolve intermingled heterogeneity. MTCM would not only make longitudinal in vivo surveillance possible but also enable imaging-informed selective biopsies. The future challenges of applying MTCM lie in the gap between research experiments and clinical practice. Unlike highquality data in well-designed research studies, clinical data are usually with limited spatial and/or temporal resolution, accompanied by higher noise level. Lower spatial resolution results in less pure-volume pixels and thus reduces the accuracy of MTCM. So far we have tested MTCM method on DCE-MRI data, dynamic contrast-enhanced optical imaging data, and dynamic PET imaging data, acquired from both human tissue/organ and whole-body mouse model. Theoretically, the MTCM method can produce confident estimation on any ‘dynamic contrast-enhanced’ imaging data with sufficient quality.