In the past decade, research in cancer biology has provided new insights into the tumor-promoting functions of stroma in breast cancer. In the presence of tumor cells, stroma produces critical signals to drive proliferation, angiogenesis and motility while suppressing apoptosis 
. In the absence of pre-existing tumor cells, stroma can also acquire genomic changes to stimulate the transformation of adjacent cells and to ultimately facilitate malignancy 
. These indicators of activated stroma may give rise to characteristic morphological features such as change in cellularity and/or vascularity that may be assessed by image-based tools. Therefore, it is important to develop new imaging methodology to assess stromal tissues before the manifestation of tumor growth and relapse. In fact, mammographically detected breast density 
and high water content measured by MRI 
are thought to reflect increased cellularity from proliferation of epithelium and stroma. Moreover, dynamic techniques using contrast-enhanced MRI further provide quantitative parameters such as peak contrast-enhancement, ktrans
and signal enhancement ratio (SER), that are correlated to intratumoral microvessel density and vascular endothelial growth factor (VEGF) expression 
. We reasoned that these imaging parameters can be used to characterize the extratumoral morphologic features of stroma and may convey information about clinical outcome.
In a previous study, SER analysis of normal-appearing breast stroma was manually performed by defining and placing regions of interest (ROI) radially from the tumor edge 
. In the current study, the semi-automated iterative segmentation technique and tumor proximity map has enabled us to calculate stromal enhancement values more precisely at specific distances around the tumor. This methodology can be applied to any registered functional image, such as PE, SER, apparent diffusion coefficient (ADC) or fractional anisotropy (FA), making it a versatile and robust method for stromal characterization.
In this cohort of 63 patients, the univariate Cox analysis showed a statistically significant association with RFS for initial tumor volume and diameter measured by MRI. A 1.7% increase in hazard of recurrence was found for each unit volume increase. Although 1.7% increase appears to be small, it can be clinically significant when a unit change in the predictor is small compared to the spread in the population; the hazard of recurrence increases exponentially (is compounded) with each additional unit change in the predictor and therefore a realistic 10 cm3
difference in tumor size between two patients would have a clinically important effect on their relative hazard (18% increase for the larger tumor size). This interpretation agrees with that of Gray 
who determined a low hazard ratio of longest diameter for predicting survival, but which was nevertheless clinically important for meaningful diameter differences.
In assessing the tumor periphery region, signal enhancement was dominated by the permeability of the gadolinium contrast agent through capillary walls in angiogenic vasculature 
measured by PE. Previously, it has shown that an increased PE in the tumor periphery was associated with increased microvessel density 
. Based on these findings, we reasoned that this measurement may be also responsive to chemotherapy. In the non-recurrent group, PE values immediately outside of the tumor (0–5 mm) after an initial exposure to chemotherapy (at V2) were estimated lower with a reduction of 3.91, 95% CI (−0.36, 8.73), p
0.07) than that at V1, possibly reflecting the inhibitory effect of chemotherapy at the tumor periphery angiogenesis.
While PE measures signal enhancement through vascular permeability of the contrast agent, SER assesses the contrast washout kinetics from the tissue. In this study, although the increasing radial trend in SER was not statistically significant, a coherent pattern of increasing estimated SER can be visualized in the non-recurrent cohort at V2 (red trend line in ). This trend was in agreement with previous findings that higher global stromal SER values (> 0.7) after one cycle of chemotherapy (V2) were associated with reduced risk of recurrence 
To characterize an overall stroma enhancement, the global PE and SER values averaged over the range of 5 to 40 mm was analyzed. A statistically significant difference between recurrent and non-recurrent groups was found in global PE at V2 (8.24, 95% CI (2.93, 12.27), p
0.02) and global SER at V1 (−0.081, 95% CI (−0.157, −0.002), p
0.04), indicating that these global measurements may reflect the overall difference of tissue biology in recurrent and non-recurrent patients and their response to therapy. In the Cox proportional hazards model for RFS, global SER at V1 was found to be a statistically significant predictor of RFS (p
0.03): for every unit decrease in SER, the estimated hazard was increased by a factor of over 17. Although the confidence interval was very wide, this finding further supports the observation of increasing radial trend of stromal SER values in non-recurrent patients, as well as previous findings of higher stromal SER in association with reduced risk of recurrence. The higher stromal SER may reflect greater microvessel density to facilitate delivery of chemotherapeutic agents, resulting in greater efficacy and reduced risk of recurrence.
The current study population is limited to women with advanced stage disease who received neoadjuvant chemotherapy. Another limitation is that there was only a small subset of the recurrent group with both MRI scans at V1 and V2. These limitations impact our ability to generalize our findings to patients with better prognosis. We recognized that our survival analyses may be influenced by the size of the measurable stromal tissue. To address this, we performed a sensitivity analysis via Cox proportional hazard modeling by incorporating stromal size as an additional covariate in each of the survival models for global PE and SER at V1 and V2. Stromal size was not a statistically significant covariate in any of the models and moreover, did not affect the general relationship between SER at V1 and recurrence. While these results do not prove that stromal size has no predictive value, they indicate that the predictive power of stromal size is limited and in particular does not appear to diminish the predictive effect of SER at V1. The current segmentation technique may pose an issue of partial volume that can be addressed by implementing an erosion correction in the future. The current study has demonstrated a new robust image-based tool for stromal characterization. Our continued efforts to refine proximity analysis of PE and SER, as well as extending the proximity mapping methodology to study the contralateral breast with additional descriptors such as ADC 
or FA 
measurements may provide additional predictive stromal imaging characteristics that can be further developed for individualized treatment intervention.