Apparent diffusion coefficient histogram in breast cancer brain metastases may predict their biological subtype and progression

Our aims for this study were to investigate the relationship between diffusion weighted image (DWI) parameters of brain metastases (BMs) and biological markers of breast cancer, and moreover, to assess whether DWI parameters accurately predict patient outcomes. DWI data for 34 patients with BMs from breast cancer were retrospectively reviewed. Apparent diffusion coefficient (ADC) histogram parameters were calculated from all measurable BMs. Two region of interest (ROI) methods are used for the analysis: from the largest BM or from all measurable BMs per one patient. ADC histogram parameters were compared between positive and negative groups depending on ER/PR and HER2 statuses. Overall survival analysis after BM (OSBM) and BM-specific progression-free survival (BMPFS) was analyzed with ADC parameters. Regardless of ROI methods, 25th percentile of ADC histogram was significantly lower in the ER/PR-positive group than in the ER/PR-negative group (P < 0.05). Using ROIs from all measurable BMs, Peak location, 50th percentile, 75th percentile, and mean value of ADC histogram were also significantly lower in the ER/PR-positive group than in the ER/PR-negative group (P < 0.05). However, there was no significant difference between HER2-postive and negative group. On univariate analysis, using ROIs from all measurable BMs, lower 25th percentile, 50th percentile and mean of ADC were significant predictors for poor BMPFS. ADC histogram analysis may have a prognostic value over ER/PR status as well as BMPFS.

Image postprocessing and analysis. We regarded a BM as measurable when its volume is more than 100 mm 3 , because with the volume of less than 100 mm 3 , it is difficult to draw ROIs exactly and co register to ADC space correctly. Regions of interest (ROIs) were drawn on each tumor section on contrast-enhanced T1-weighted images using a free open-source toolkit, ITK-SNAP (www.itksnap.org) 17 . ROI drawings were not performed for non-measurable BMs. The ROI masks were automatically segmented with intensity thresholds; incomplete regions of the entire enhancing tumor were manually corrected. ROIs were co-registered to ADC maps via affine transformation with normalized mutual information as a cost function 18 . ADC histograms were generated with a bin of 1 × 10 −5 mm 2 /sec and a range of 10-3000 × 10 −6 mm 2 /sec. We considered ADC values <10 × 10 −6 mm 2 / sec to be artifacts and values >3000 × 10 −6 mm 2 /sec to be cystic portions. ADC histogram parameters (peak location, 25th and 75th percentile values, median, mean, and standard deviation) were calculated from ROIs that were overlaid on ADC maps. A representative case is presented in Fig. 1.

Statistical analysis.
Overall, we used two ROI methods for our study: First one is to use a ROI of largest BM from one patients. In this case, we used a total of 34 ROIs for analysis. Second one is to use ROIs from all measurable BMs. In this case, we used a total of 85 ROIs for analysis. ADC histogram parameters were compared between positive and negative groups depending on ER/PR and HER2 statuses, using a two-sample t-test. ADC variables were dichotomized into two groups by median and univariate survival analyses were performed to identify ADC variables to stratify OSBM and BMPFS, using a log-rank test.
Prediction of overall survival after brain metastasis and brain metastasis-specific progression-free survival. On univariate analysis with ROI of largest BM, ADC variables were not significant prognostic factors for OSBM and BMPFS. However, with ROIs from all measurable BMs, peak location, 25th percentile, 50th percentile, and mean of ADC histogram were significant prognostic factors for BMPFS but not for OSBM, p < 0.05, Table 4). Lower ADC variables showed poor BMPFS (Fig. 2).

Discussion
In this study, we tested the hypothesis that ADC histogram analysis of BMs can accurately predict the biological subtypes of breast cancer, and patients' outcomes. Our results indicated that ER/PR-positive patients have significantly lower 25th percentile of ADC values in BM, compared with ER/PR-negative patients, regardless of ROI methods (ROIs from largest BMs or ROIs from all measurable BMs). Using ROIs from all measurable BMs, peak location, 50th percentile, 75th percentile and mean of ADC were also significantly lower in ER/PR-positive patients than in ER/PR negative-patients. However, ADC variables are not correlated with HER2. Using ROIs from all measurable BMs, peak location, 25th percentile, 50th percentile, and mean of ADC significantly predicted BMPFS, thus they could be a potential prognostic biomarker for BMPFS.   20 . However, to date, no research has examined the relationship between ADC values and biological features of BMs from primary breast cancer. In the current study, all ADC variables in BMs from breast cancer revealed a decreasing trend in the ER/PR-positive group, compared with the ER/PR-negative group. This result is consistent with previous results in which the median ADC values of primary breast cancer were significantly lower in the ER-positive group than in the ER-negative group 21,22 . This phenomenon can be explained as follows: the ADC value is affected by the molecular diffusion of water, as well as by perfusion 23,24 . Studies using experimental models have shown that ERs inhibit the angiogenic pathway and induce a decrease in perfusion, thus affecting the ADC value 25 . Another noticeable thing   Table 2. Comparison of apparent diffusion coefficient (ADC) variables from volume of interest (VOI) of largest brain metastasis according to human epidermal growth receptor 2 (HER2) and estrogen receptor/progesterone receptor (ER/PR) status.
SCiEntifiC REPORts | (2018) 8:9947 | DOI:10.1038/s41598-018-28315-y in our results is that 25th percentile of ADC shows a significant difference between ER/PR-positive and negative groups regardless of ROI methods. There is a dilemma in choosing ROI method: Using a single representative BM or all measurable BMs from one patient. Either way has flaws, in case of first method, it is hard to select the representative lesion. For the second method, an oversampling issue is raised. However, 25 th percentile of ADC is not dependent on ROI method and might be useful to differentiate two groups. Our results showed that ADC variables are not correlated with the HER2 status of BMs from primary breast cancer. There has been a controversy regarding this issue in primary breast cancer. Jeh et al. showed that ADC variables of primary breast cancer were significantly lower in the HER2-positive group, compared with the HER2-negative group 26 Table 4. Survivals in breast cancers with brain metastases depending on apparent diffusion coefficient (ADC) histogram analysis. BMPFS, Brain metastasis-specific progression-free survival; OSBM, overall survival analysis after brain metastasis. Some studies have postulated a prognostic value for ADC values in primary breast cancer. Nakajo et al. examined 44 breast cancers, and concluded that a low ADC value is significantly correlated with poor prognosis 27 . Mori et al. investigated 86 patients with luminal type breast cancer and showed that ADC values are correlated with the Ki-67 labeling index, which is a significant prognostic factor 28 . For BMs, Lee et al. evaluated the effect of stereotactic radiosurgery on BMs with ADC maps; they found that increased ADC values are indicators of good tumor control. However, because their primary tumor origins were heterogenous, it is difficult to apply their results to BMs that originate from primary breast cancer. In our study, 25 th percentile, 50 th percentile, and mean of ADC show potentials to predict BMPFS, specifically using ROIs from all measurable BMs.
Our study has a limitation, the number of our cohort is small for multivariate analysis. We could not verified our independent predictability of our ADC variables. However, our cohort is rather homogenous because patients had been recruited in same institution for a long period (7 years). Also, clinical characteristics between two groups (HER2 positive vs negative, ER/PR positive vs negative) are not different. Thus, our results may serve as a cornerstone for future studies with a larger population to validate and extend these results.
In conclusion, we demonstrated that ADC variables of BMs in breast cancer are significantly lower in ER/ PR-positive patients than in ER/PR-negative patients. Specifically, the 25th percentile ADC value are consistently different between two groups regardless of ROI methods. Also, ADC variables of BMs may be a prognostic indicator for BMPFS of breast cancer but these are necessary to be verified in future study with large cohort.
Data availability. All data generated or analyzed during this study are included in this published article and its Supplementary Information files.