Dynamic Contrast-Enhanced (DCE) MRI has been widely used to assess breast tumor microvasculature. Temporal resolution has reduced from above a minute, which are multiple high spatial resolution post-contrast acquisitions, to about 4s, which is considered very fast. Analysis methods range from qualitative visual inspection of signal intensity time curves, to semi-quantitative parameters such as time to peak and uptake slope of intensity or tracer concentration time curves (C(t)), and to quantitative tracer kinetics (TK) analysis. Compartmental models are more commonly used for quantitative analysis while more realistic distributed parameter (DP) models¹ are less commonly used. Here, we report on high temporal resolution DCE-MRI analyzed with both types of TK models in a clinical trial that assess early chemotherapy response with or without anti-angiogenesis pre-treatment, in breast cancer.

Sixty-four newly diagnosed breast cancer patients in an on-going clinical trial who were scheduled to receive neo-adjuvant chemotherapy (doxorubicin/cyclophosphamide (AC)), were randomly assigned to receive an additional low-dose (oral Sunitinib (Su) at 12.5mg/d) short-duration (7 days) anti-angiogenesis pre-treatment (ARM-B) or not (ARM-A)². DCE-MRI data were acquired at these time points: before anti-angiogenic pre-treatment using Sunitinib, before and after first cycle of chemotherapy.

MRI scans were performed on a whole-body 3T MRI (Magnetom Trio; Siemens, Germany) with a seven-channel breast receiver coil and an additional surface coil placed on their back for better imaging of the descending aorta. Axial DCE-MRI were scanned in the prone position with a spoiled 3D FLASH acquisition (TR=4ms, flip angle (FA)=15°, matrix=128×128, 16x4mm slices to cover whole tumor, temporal resolution=2.4s/f). T1 maps were acquired with variable FA (FAs=5°/13°/20°, TR=20ms) for computation of tracer concentrations. B1 maps were acquired using the system RF sequence for B1 correction.

Both the modified Tofts (mT) and 2-comparment DP models¹ were used to analyze the DCE data. AIF is measured from the fourth inferior slice at the descending aorta to reduce in-flow effects. A physiological based AIF model³ was fitted to the AIF data and later used in the fitting of the tissue C(t). This allows AIF arrival time to be adjusted freely while fixing the TK impulse response function (IRF) to always start at time 0 to overcome the issue with discontinuity at the bolus arrival time. Discontinuity issue at the mean vascular transit time (t_p in Figure 1 right) the DP model is handled by up-sampling of AIF and IRF by 10 times, before applying the correction method by Koh et al.⁴ Using an analytical AIF form further smooths the search surface during curve fitting. All processing were performed on MatLab, where the expansion series form of the DP model⁵ was implemented for faster computation.

Figure 1 shows typical AIF and tissue C(t). The high temporal resolution clearly captured the first bolus and a second smaller peak in the AIF. Our AIF model managed to fit the AIF data appropriately. We notice the scaled AIF in the mT model did not match the initial part of many tissue C(t) (Figure 1 centre). The K^trans and permeability surface-area product (PS) maps (Figure 2) were similar but differed in values. Table 1 shows the results from 48 patients who had completed all scans. All parameters did not show any significant difference in ARM-A. In ARM-B, fractional plasma volume (v_p) and fractional extravascular extracellular volume (v_e) in the mT model showed more changes than v_p and v_e in the DP model (circles in Table 1). However, DP permeability measures (PS, K^trans (see below) and first pass extraction ratio (E)) showed significant changes at post-AC while mT K^trans did not. Perfusion (F) measured by DP showed no significant changes between all scans. The K^trans in DP model is defined as E×F.We have performed high temporal resolution DCE MRI of breast cancer and analyzed the data with more realistic DP and the commonly used mT models. The initial AIF features captured in our data illustrated a misfit issue (Figure 1 centre) in the mT model. Although permeability measures from both models showed similar trend, mT K^trans and v_e were much higher than those of DP model. Since mT K^trans also contains perfusion contributions, and there was no significant changes in the perfusion as measured by DP F, we suspect this could have muddled the values in mT K^trans to not reflecting significant changes in permeability while all three permeability measures of DP model have.The difference between mT and DP models is noticed in analyzing high temporal resolution. More realistic distributed parameter models might be better in analyzing high temporal resolution DCE MRI data.