Prediction of response to preoperative or ‘neoadjuvant’ chemotherapy (NAC) can help guide treatment strategies for patients with triple-negative breast cancer (TNBC), a highly aggressive subtype of breast cancer. Breast magnetic resonance imaging (MRI) can provide noninvasive measurements of the microenvironment across a tumor volume. We hypothesize that pre-treatment measurements from dynamic contrast-enhanced (DCE-) MRI reflecting tumor perfusion and vascular function are predictive of NAC response for TNBC patients. Women with TNBC who underwent pre-treatment MRI and NAC at our institution (2005-2019) were retrospectively identified. DCE-MRI was acquired at 2, 5, and 8 minutes after contrast injection. From DCE-MRI, whole tumor contrast kinetics measures including functional tumor volume (FTV), percent enhancement (PE) at 2 mins post-contrast and signal enhancement ratio (SER) were calculated, and hotspot measures of peak PE and peak SER (representing the highest mean PE and SER, respectively, for 3?3 voxel subregions) were determined. Imaging measurements were compared between those with complete pathologic response (pCR; no residual cancer present in the breast at surgery) and non-pCR patients with a two-tailed Student’s t-Test (p<0.05 considered significant). 95 women (median age: 49, range: 30-79 years) with TNBC were evaluated, of which 29 (31%) achieved pCR. FTV was significantly higher in non-pCR patients (21.1±28.1 cc) compared to pCR patients (8.6±11.3 cc, p<0.01). Peak SER was also higher in non-pCR patients (1.8±0.3) compared to pCR patients (1.7±0.3), trending toward significance (p=0.06). No significant differences between groups were observed in peak PE measures. Patients with lower pre-treatment tumor FTV and peak SER on DCE-MRI were more likely to achieve pCR after standard NAC. These findings indicate that baseline DCE-MRI measurements may help predict response and assist in optimizing treatment plans for TNBC patients, such as selecting more aggressive regimens incorporating immune checkpoint inhibitors or other novel agents in predicted non-responders.

Women with dense breasts have increased amounts of fibroglandular tissue (FGT) and are at higher risk of developing breast cancer. Quantitative measurement of FGT from magnetic resonance imaging (MRI) could provide more robust measurement of density, supplanting conventional qualitative radiologist assessments. Current quantitative methods involve manual selection of a signal intensity threshold, which can be time consuming and subjective. Fuzzy c-means (FCM) clustering is an automated approach to tissue segmentation, offering a reproducible process for quantifying FGT volume. The aim of this study is to evaluate the efficacy of the FCM clustering in identifying FGT compared to manual thresholding. Women (N=10) who underwent screening breast MRIs at our institution were evaluated in this preliminary study. Fat-suppressed T₁-weighted pre-contrast images acquired as part of their clinical breast MRI exams were used for FGT segmentation. Prior to segmentation, I cropped the images to include only the breast. FGT was then segmented two ways, 1) manually, using a signal intensity threshold that I chose and adjusted and 2) automatically, using existing lab software for FCM clustering. The Sørensen-Dice similarity coefficient was calculated between the manual and automatic segmentations for each patient to determine the degree of overlap. The concordance correlation coefficient (CCC) was calculated between automatic and manual segmentation volumes across the whole data set. Across the 10 patients, an average (± standard deviation) Dice coefficient of 0.81±0.04 was observed, indicating good spatial agreement between the manual and automatic segmentations. The CCC between the FGT volume from manual and automated segmentation was 0.89, demonstrating high correlation in volume estimates between the two methods. Fuzzy c-means clustering was determined to be an effective and efficient method of FGT segmentation in breast MRI data. Future work will evaluate the application of this technique in assessment of background parenchymal enhancement, a clinical marker of cancer risk.