Neoadjuvant chemotherapy (NAC) is increasingly used in breast cancer because it enables breast-conserving surgery in women who traditionally required a mastectomy(1, 2). The goal of NAC is a pathologic complete response (pCR) defined as the absence of any residual in situ or invasive cancer(3). The National Comprehensive Cancer Network (NCCN) guidelines currently recommend a breast MRI before and after NAC. Multiparametric MRI has the highest diagnostic accuracy for pCR (1,4). Pathologic complete response (pCR), an intermediate end point, serves as a biomarker for improved disease free and overall survival(3, 5). Computer extracted MRI features can generate quantitative metrics but no imaging biomarker has been validated to accurately predict or diagnose a pCR post NAC. The purpose of this study was to extract MR image features of breast cancers on multiparametric sequences pre and post NAC and determine their ability to predict a pCR.

This retrospective study received institutional review board approval and need for informed consent waived. Clinical breast MRIs pre and post NAC were examined in 20 consecutive patients with invasive ductal carcinomas in 2013-2014. One patient had a tumor in both her breasts. A total of 21 tumors were analyzed as one patient had synchronous bilateral breast cancer. MRI examinations were performed with a 3.0-T whole-body MRI unit (GE Medical Systems) equipped with a dedicated 8 or 16-channel breast coil. The standard axial clinical protocol was analyzed and included: DWI (b= 0, 600, 1000); T2-weighted (w) fat-suppressed (FS); T1-w; T1-w FS before intravenous contrast; 5 time point high-temporal (12 seconds) T1-w FS post-contrast; and three time point high-spatial (130 seconds) T1-w FS post-contrast sequences. These 8 different sequences were analyzed and henceforth are referred to as: DWI, T2, T1, T1 FS Pre, High Temporal PC, T1 FS First PC, T1 FS Second PC and T1 FS Third PC, respectively. Gadolinium was injected intravenously as a bolus of 0.1mmol/kg at a rate of 2ml/s by an MR- compatible power injector.

The tumors were segmented volumetrically using interactive Grow-Cut method available in 3DSlicer (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3586703/). From the segmented regions, 5 Haralick features were computed voxel-wise: contrast, correlation, energy, entropy and homogeneity generating a total of 40 different features(6). A neighborhood of 5x5x5 pixels and 128 bins were used in the computation. From the computed texture maps, the mean values of the textures inside the volume-of-interest (VOI) were used for the analysis. Difference of all of the aforementioned features computed from the pre and post-treatment MRI were computed and statistical significance of the different features between patients with pCR and no-pCR at time of definitive surgery were evaluated by means of unpaired two-sided T-test at a 95% confidence level.

Of the 21 invasive tumors, 4/21 (19%) had a pCR and 17/21 (81%) had no-PCR. When using just the texture features from pre-NAC MR images, out of the forty texture features, T1 contrast texture (P=0.02) and T1 homogeneity (P=0.04) were significantly different between pCR and no-pCR. There were differences between MR image texture features pre and post-NAC. The T1 contrast texture (P=0.03), T1 homogeneity (P=0.05), High temporal PC correlation (P=0.02), High temporal PC homogeneity (P=0.03) and T1 FS First PC contrast texture (P=0.03) were significantly different between pCR and no-pCR (Figure 1, Figure 2 and Figure 3). In general, pCR tumors on pre-NAC MRI were more homogenous compared to no-pCR tumors evidenced by lower contrast texture (pCR=0.35; no pCR=0.49) and higher homogeneity (pCR=0.38; no-pCR=0.32) on T1. A similar result was observed when measuring the difference in the texture values between the pre and post NAC MRI. Tumors with a pCR showed a greater change in the MRI texture values pre and post NAC compared to no-pCR for the T1 contrast texture (pCR= 0.06; no-pCR=0.002), T1 homogeneity (pCR= 0.08; no-pCR = 0.02), high temporal PC correlation (pCR= -0.26; no-pCR=0.11), and high temporal PC homogeneity pCR=-0.11; no-pCR=0.14). We found a difference in the MRI tumor texture features pre and post NAC that were significantly different between pCR and no-pCR. Patients with pCR in general had more homogeneous tumors pre-NAC MRI and tended to show a larger change in the texture values post-NAC compared to patients with no-pCR. These quantitative metrics demonstrate MRI evident changes in tumor microenvironment post NAC and are biomarkers for pCR and consequently disease free and overall survival. Larger studies are planned to validate these findings, which may provide further insight into tumor biology and personalized care as it pertains to treatment monitoring, response and outcome.