MRI primary breast lesion volume has been shown to be a strong predictor in the response to chemotherapy for invasive breast cancer. However, the prediction accuracy remains low. In this study, we investigated the feasibility of using lymph node volume and signal enhancement ratio as predictors of the chemotherapy response. The result shows that the lymph nodes signal-enhancement ratio is a stronger predictor than lymph node volume and primary in-breast lesion volume.
Primary tumor response monitoring has led to a greater use of neoadjuvant chemotherapy for women with breast cancer. In the Receiver Operating Characteristic (ROC) and Area Under Curve (AUC) analysis of a recent breast cancer treatment clinical trial (I-SPY 1), MRI primary breast lesion volume and signal enhancement ratio (SER) were found to be strong predictors of pathological response (AUC of ~0.6) 1. While this is encouraging, this prediction accuracy is low. We proposed to use lymph node volumes and SERs as pathological response predictors.
Thus, in this study, we performed similar ROC analysis to assess the predictive values of MRI lymph node characteristics in the I-SPY 1 data set. The MRI-derived pre-treatment and post-treatment lymph node volumes and SER after contrast were analyzed. Single variable and multivariable analyses were also performed on lymph node volumes and SERs in the pre- and post-chemotherapy.
The I-SPY 1 MRI of 212 breast cancer patients undergoing neoadjuvant therapy with an anthracycline-cyclophosphamide regimen alone or followed by taxane was used1-4. Among this data set, we included 27 patients (9 pathologic complete responders and 18 incomplete or non-responders) with readily available T2-weighted MRI and DCE data to evaluate the prediction of pathological chemotherapy response.
The lymph node volumes were manually contoured from T2-weighted MRI and DCE MRI at two time points: pre- and post-treatment. The primary quantitative measurements were lymph node volume estimates and the SERs, defined as (S1 − S0)/(S2 − S0), where S0, S1, and S2 represent the signal intensities on the pre-contrast, early post-contrast, and late post-contrast images, respectively. The largest diameter (LD) of breast lesions included in the I-SPY open data set was performed as a reference predictor. Pathological complete response (pCR) score of the primary breast tumor was used as ground truth.
We used pre-treatment and post-treatment lymph node volume and SER to predict the pathological response. Then the change rate (CR) was calculated as another predictor. CR was defined as (Ppost-treatment − Ppre-treatment)/Ppre-treatment, where P represents the SER or lymph node volume. In addition, a multi-parametric predictor was generated, evenly weighting SER and lymph node volume. ROC curve was generated to visualize the tradeoffs between sensitivity and specificity and the AUC was computed as a quantitative value to evaluate the prediction.
For prediction of pathologic response following neoadjuvant chemotherapy, the AUC for lymph node lesion volume was 0.51, lymph node SER was 0.71, compared to primary in-breast lesion volume of 0.53.
In a previous I-SPY-1 study1 with a larger sample size (N = 216), the AUC of primary in-breast lesion volume for predicting pCR was ~0.6, compared to our in-breast lesion volume AUC of 0.53 (which used a subset of the I-SPY-1 data). Our results are in general agreement with previous finding on in-breast lesion volume AUC. Our ROC results showed that the lymph nodes signal-enhancement ratio is a stronger predictor than lymph node volume and primary in-breast lesion volume.
When clinical examination was included in their previously analysis, the AUC increased to 0.73. With the inclusion of race and age, multivariate analysis yielded AUCs of 0.84 for the pre-treatment time point1. Our sample sizes were too small to include multivariate analysis at this time.
One limitation of this study was that the low spatial resolution in the T2-weighted MRI and DCE images caused difficulties contouring lymph nodes, which may have caused uncertainties in the lymph node volumes. In addition, the DCE images obtained at three phases would preclude the dynamic pharmacokinetic modeling to extract the tumor blood volume. Our small sample size is small. Despite the small sample sizes, our findings were statistically significant. However, this finding needs to be repeated on a larger sample size. Future studies will incorporate multiple parameters into ROC analysis. Despite these limitations, our ROC results on lymph node SER compared favorably to lymph node lesion volume and in-breast lesion volume.
1. Hylton NM et al., Locally advanced breast cancer: MR imaging for prediction of response to neoadjuvant chemotherapy--results from ACRIN 6657/I-SPY TRIAL. Radiology 263, 663-672 (2012).
2. Hylton NM et al., Neoadjuvant Chemotherapy for Breast Cancer: Functional Tumor Volume by MR Imaging Predicts Recurrence-free Survival-Results from the ACRIN 6657/CALGB 150007 I-SPY 1 TRIAL. Radiology 279, 44-55 (2016).
3. Clark K et al., The Cancer Imaging Archive (TCIA): maintaining and operating a public information repository. J Digit Imaging 26, 1045-1057 (2013).
4 Newitt D, Hylton N, on behalf of the I-SPY 1 Network and ACRIN 6657 Trial Team. (2016). Multi-center breast DCE-MRI data and segmentations from patients in the I-SPY 1/ACRIN 6657 trials. The Cancer Imaging Archive. http://doi.org/10.7937/K9/TCIA.2016.HdHpgJLK