Quantitative Changes in Intratumoral Habitats on MRI Correlate With Pathologic Response in Early-stage ER/PR+ HER2− Breast Cancer Treated With Preoperative Stereotactic Ablative Body Radiotherapy

Abstract

Objective:

To quantitatively evaluate intratumoral habitats on dynamic contrast-enhanced (DCE) breast MRI to predict pathologic breast cancer response to stereotactic ablative body radiotherapy (SABR).

Methods:

Participants underwent SABR treatment (28.5 Gy x3), baseline and post-SABR MRI, and breast-conserving surgery for ER/PR+ HER2− breast cancer. MRI analysis was performed on DCE T1-weighted images. MRI voxels were assigned eight habitats based on high (H) or low (L) maximum enhancement and the sequentially numbered dynamic sequence of maximum enhancement (H1−4, L1−4). MRI response was analyzed by percent tumor volume remaining (%VR = volume post-SABR/volume pre-SABR), and percent habitat makeup (%HM of habitat X = habitat X voxels/total voxels in the segmented volume). These were correlated with percent tumor bed cellularity (%TC) for pathologic response.

Results:

Sixteen patients completed the trial. The %TC ranged 20%–80%. MRI %VR demonstrated strong correlations with %TC (Pearson R = 0.7–0.89). Pre-SABR tumor %HMs differed significantly from whole breasts (P = 0.005 to <0.00001). Post-SABR %HM of tumor habitat H4 demonstrated the largest change, increasing 13% (P = 0.039). Conversely, combined %HM for H1−3 decreased 17% (P = 0.006). This change correlated with %TC (P < 0.00001) and distinguished pathologic partial responders (≤70 %TC) from nonresponders with 94% accuracy, 93% sensitivity, 100% specificity, 100% positive predictive value, and 67% negative predictive value.

Conclusion:

In patients undergoing preoperative SABR treatment for ER/PR+ HER2− breast cancer, quantitative MRI habitat analysis of %VR and %HM change correlates with pathologic response.

Introduction

MRI is the most accurate method for evaluating breast cancer response to neoadjuvant therapy (1). For instance, MRI demonstrated better accuracy for pathologic complete response (pCR) and residual disease prediction compared to clinical examination and mammography in a meta-analysis of 44 studies (2). The most widely used assessment to evaluate tumor response on MRI is diameter or volume measurements following Response Evaluation Criteria in Solid Tumors 1.1 guidelines as well as tumor morphological features and qualitative enhancement kinetic curves (1,3,4). However, there are limitations, as MRI sensitivity for depiction of residual disease ranges 63%–88% with specificity 54%–91% in meta-analyses (5).

MRI limitations in evaluating residual disease is particularly an issue in ER/PR+ HER2− breast cancers. They are more likely to be underestimated on post-treatment MRI than other breast cancer phenotypes (6,7). Given the limitations in evaluating these tumors, we hypothesized that quantitative microhabitat analysis within the tumor, beyond the standard volumetric and kinetic curve analysis, could demonstrate additional benefits in predicting pathologic response. The Stereotactic Ablative Body Radiotherapy (SABR) trial (NCT03137693) of neoadjuvant radiation treatment for early-stage ER/PR+ HER2− breast cancer completed at our institution was selected for evaluation of a perfusion-based quantitative breast MRI analysis because of its uniform tumor set. Stereotactic ablative body radiotherapy is a neoadjuvant partial breast irradiation (PBI) technique that uses radiation beams of small cross-sections targeting a tumor volume from multiple different positions. Phase I and II studies to date show SABR is a feasible, well-tolerated, and reliable PBI approach in the neoadjuvant and adjuvant treatment of breast cancer (8–10). Studies also show correlative response on breast MRI. In a preliminary study by Wang and colleagues, 15 patients with breast cancer were imaged preoperatively by diffusion-weighted MRI (DW-MRI) and dynamic contrast-enhanced MRI (DCE-MRI) approximately 1 week before and 1 week after 15–21 Gy SABR. The model-free parameter initial area under the concentration curve and the Tofts model parameter v_e were found to be significantly increased following SABR, with the increase greater at higher SABR doses.

Standard reported MRI response in the SABR study at our institution has been published, demonstrating correlation of reported cubic volume change as well as Breast Imaging Reporting and Data System (BI-RADS) (11) descriptor change with pathologic response (12). We hypothesized that additional quantitative measurements of tumor response would demonstrate further correlation with pathologic response. Indeed, analysis of image texture features within objectively defined “intratumoral habitats” has been used to predict molecular characteristics of breast cancer, pathologic response, and recurrence-free survival following neoadjuvant chemotherapy (NAC) for breast cancer (13–15). A technique using multispectral habitat analysis was developed at our institution and previously demonstrated utility in the analysis of glioblastoma multiforme (16). For this study, we applied this technique to analyze breast tumors in the SABR trial to determine quantitative results predictive of pathologic response and determine utility over the standard previously reported size, descriptors, and kinetic curve results.

Methods

Study Design and Patient Population

The images analyzed in this study were acquired on patients enrolled in a Health Insurance Portability and Accountability Act–compliant, Institutional Review Board–approved, phase 2 trial of neoadjuvant SABR at a single institution between November of 2017 and October of 2019 (https://clinicaltrials.gov/ct2/show/NCT03137693). Written informed consent was obtained for enrollment, and enrollment was capped at 21 patients. Inclusion criteria for the trial were women 50 years or older with unifocal, invasive adenocarcinoma of the breast, 2 cm or less in size based on MRI, clinically node-negative (cN0), ER+ HER2−, with target lesions >10 mm from the skin based on MRI. These patients had to be clinically and radiologically lymph node negative, have no history of other invasive malignancy in the prior five years, and no history of ipsilateral breast or thoracic radiotherapy. Patients with BRCA1 or BRCA2 mutations were excluded. Inclusion and exclusion criteria were guided by the American Society for Radiation Oncology updated consensus statement for accelerated partial breast irradiation (17).

Treatment

CT simulation was performed on each patient using MRI co-registration for tumor delineation. Next, patients underwent preoperative SABR at 28.5 Gy given in three fractions of 9.5 Gy on different days separated by no more than 48 hours. Approximately five to six weeks following SABR treatment, a follow-up MRI was performed at our institution. Breast conservation surgery was then performed on the tumor site at six to eight weeks post-SABR treatment. This was performed with the aid of image-guided localization of the residual tumor with biopsy marker.

Pathology Analysis

Pathology reports from the initial core biopsy were evaluated for histopathology and receptor status. Final surgical pathology reports were evaluated for histopathology, sentinel lymph node pathology, percent tumor cellularity (%TC, percentage of residual tumor within the tumor bed), and percent tumor infiltrating lymphocytes (TILs). All reports were dictated by the study pathologist, a subspecialized breast pathologist. Finally, a combined %TC and %TILs score was calculated. Following the method outlined by Nuciforo and colleagues, “CelTIL” scores were calculated using the formula CelTIL = −0.8x%TC + 1.3x%TILs (18). Complete pathologic response (pCR) was defined as no residual invasive tumor in the surgical bed, partial pathologic response (pPR) as %TC ≤70%, and nonresponders (pNR) as %TC >70%.

Surgical specimens were examined fresh, x-rayed, and sliced at 5-mm intervals for intraoperative margin evaluation consultation. Fixation in 10% buffered formalin was performed following national guidelines, and the entire area of interest was mapped and submitted by a pathologist assistant. For final histological evaluation, the Residual Cancer Burden algorithm was used by the study pathologist. This includes tumor bed size, percentage cancer cellularity, percentage in situ disease, and the presence and extent of involvement of lymph nodes (19). For the evaluation of TILs, the recommendations of the International TILs Working Group were followed (20).

MRI Inclusion

All patients in the SABR treatment trial were required to have a pretreatment baseline MRI performed at our institution or an outside institution. All post-SABR treatment MRIs were performed at our institution. The Institutional Review Board protocol was amended and approved to include retrospective quantitative MRI analysis. For inclusion in this image analysis study, patients with baseline MRIs performed at an outside institution were excluded.

MRI Protocol

Breast MRIs were performed with participants prone using a dedicated GE 1.5T Optima 450w MRI scanner (GE Healthcare, Chicago, IL) with Invivo Sentinelle (Dunlee, Hamburg, Germany) 16-channel breast array coil. Standard imaging protocol was used with axial pre-contrast, non-fat-saturated images (matrix size 512 × 512, field-of-view [FOV] 300 × 300 mm, slice thickness 1.5 mm, no slice gap), one pre- and four post-contrast T1-weighted axial fat-suppressed 3D spoiled gradient-echo sequences (matrix size 512 × 512, FOV 300 × 300 mm, slice thickness and spacing 1.5 mm, flip angle 10), and an axial pre-contrast T2-weighted short tau inversion recovery sequence (TR = 6835 ms, TE = 24 ms, matrix size 512 × 512, FOV 300 × 300 mm, 3-mm slice thickness no gap). Gadobutrol gadolinium-based contrast (Gadovist, Bayer HealthCare Pharmaceuticals, Whippany, NJ) was power-injected at a concentration of 0.1 mmol/kg. T1 fat-saturated post-contrast scans were performed at 90-second intervals.

MRI Dimension and Descriptors Analysis

Patients who successfully completed the SABR trial with baseline and post-SABR presurgical MRIs performed at our institution were selected for MRI analysis of treatment response. MRI reports were completed by 1 of 8 subspecialized breast imaging radiologists at our institution with 3–12 years of experience in breast imaging using the BI-RADS MRI lexicon as recommended by the American College of Radiology (11). All images were analyzed using Dyncacad software (InVivo Phillips, Gainesville, FL). Tumor response as reported in the MRI reports was analyzed and has been published previously (12).

Tumor segmentation was performed on T1 fat-saturated pre-contrast-subtracted post-contrast images in the first (early) post-contrast sequence with software automation carrying this segmentation to the additional post-contrast sequences. Tumors were segmented using the Quantitative Imaging Decision Support platform (Healthmyne, Madison, WI) by drawing the long dimension and perpendicular short dimension on the slice with the largest tumor dimensions as determined by the segmenting radiologist. Using these inputs, the segmentation volume was semiautomatically calculated by the Healthmyne software, with manual corrections performed as needed within the software by the radiologist. Segmentation was performed on all post-SABR MRIs, including cases where the MRI report suggested complete MRI response, by using landmark and biopsy marker artifact correlation. In all cases, the segmenting radiologist was blinded to pathology results prior to segmentation. The segmentation radiologist was a fellowship-trained breast imaging radiologist with six years of experience as a dedicated breast radiologist (RJW).

MRI Preprocessing and Multispectral Habitat Analysis

The fat-saturated pre-contrast and four dynamic post-contrast-enhanced T1-weighted images acquired in a session were co-registered to the first post-contrast image by deformable registration using a previously described method (16). Following image registration, bilateral breast volume segmentation was accomplished using three main steps as described recently: landmark identification, chest wall and pectoralis muscle isolation, and skin removal (21). All voxels within the segmented bilateral whole breasts were subjected to habitat analysis to define eight tissue types based on the degree of maximum contrast enhancement on MRI (high or low using the Otsu method) (22) and by which of the four DCE-MRI phases maximum enhancement was achieved. The degree of contrast enhancement was defined as (I_post,j—I_pre)/I_pre, where I_pre is the pre-contrast voxel intensity and I_post,j is the post-contrast voxel intensity at sequential dynamic phase scan j (= 1, 2, 3, or 4). The eight final tissue types defined thus were: H1 (high enhancing, maximum achieved at j = 1), H2 (high enhancing, maximum achieved at j = 2), H3 (high enhancing, maximum achieved at j = 3), H4 (high enhancing, maximum achieved at j = 4), L1 (low enhancing, maximum achieved at j = 1), L2 (low enhancing, maximum achieved at j = 2), L3 (low enhancing, maximum achieved at j = 3), and L4 (low enhancing, maximum achieved at j = 4). Examples of whole breast and tumor habitat segmentation with color overlay are shown in Figure 1. Tumor habitat analysis focused on H1–4, with L1–4 considered subthreshold enhancement.

Figure 1.

71-year-old female with right breast invasive ductal carcinoma at 9 o’clock spanning 12 mm. Single slice axial T1 post-contrast fat-subtracted MRI images demonstrating whole breast (A). Habitats H1–4 and L1–4 are represented within the segmentation by individual colors (B). Tumor habitat segmentation (arrow, C) on a baseline exam with a zoomed in view (D).

Microhabitat Volume and Habitat Analysis With Correlation to Pathologic Response

Quantitative analysis was performed using two methods: percent habitat makeup (%HM) and percent volume remaining (%VR). Quantitative whole breast and tumor %HM analysis was performed by adding the number of voxels in each habitat together and dividing by the total voxels in all habitats combined (whole tumor and whole breast) to determine %HM for each habitat. Analysis was performed for individual habitats and habitat combinations. Specific H1–4 habitat combinations were guided by dynamic phase and by percent change analysis of the individual habitats. The %HM of tumors was compared to %HM of whole breasts both pre- and post-SABR to determine significant differences. Habitats and habitat combinations with the largest percent makeup change between pre- and post-SABR MRIs were then evaluated for use as a diagnostic test to differentiate pPR from pNR to calculate accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). True positives were defined as those that accurately predicted pPR, and true negatives accurately predicted pNR.

Tumor volume was calculated using 3D orthogonal cubic volume as obtained from the MRI report and by segmentation habitat voxel summation. Voxel volume was calculated for the whole tumor as well as for individual habitats and habitat combinations. For each of these volume measurements, the percent volume remaining (%VR) was calculated by dividing post-treatment volume by pretreatment volume in each subject. The %VR using the various calculation methods was correlated with %TC as reported on the final surgical specimen pathology. Next, to account for possible immune response, these habitat makeup analyses and volume analyses were correlated with TILs and also with the combined CelTIL scores.

Statistical Analysis

The %HM of the whole breasts was compared to that of the tumors using paired t-test to determine differences for individual habitats and habitat combinations. For treatment effect analysis, the tumor %HM changes on MRI were compared using paired t-test to determine significant changes. Habitat makeup changes used in diagnostic testing to determine pPR versus pNR were also used to compare to %TC means using independent means t-test calculator. Habitats and habitat combinations with the largest %HM change in the tumor between pre- and post-SABR MRIs were evaluated for use as a diagnostic test to differentiate pPR from pNR. Habitat makeup and %VR using methods outlined above were correlated with %TC and CelTIL using Pearson’s correlation coefficient. For purposes of this pilot study, P values < 0.05 were considered statistically significant. All tests were performed using Social Science Statistics software (23).

Results

A total of 20 patients completed the SABR trial. Four patients had baseline MRIs performed at an outside institution and were excluded. Therefore, 16 female patients in the SABR trial met criteria for inclusion in our study with a mean age of 66 years (range 55–74 years). Tumor size averaged 12 mm (range 6–19 mm). All tumors were ER/PR+ HER2−. Tumor cellularity in the surgical specimens ranged 20%–80% (median 30%) with no pCR (Table 1). Fourteen patients had pPR (%TC ranging 20%–50%), and two patients had pNR (80%TC for both). Post-SABR MRIs were performed an average of 39 days after SABR treatment (range 31–42 days). Surgery was performed an average of 47 days after SABR treatment (range 40–60 days) and an average of 8 days after post-SABR MRI (range 1–21 days).

Table 1.

Patient and Tumor Characteristics Pre- and Post-SABR

Characteristics	Average	Standard Deviation	Range
Age (years)	66	5	55–74
Reported initial tumor volume (mm3)	1315	1240	180–4788
Reported post-SABR tumor volume (mm3)	564	530	0–1170
Tumor cellularity (%)	38	19	20–80
Tumor infiltrating lymphocytes (%)	11	5	5–20
Normalized CelTIL scores	61	26	0–100
% tumor volume remaining by report	47	36	0–117

Abbreviations: CelTIL, Tumor Cellularity and Tumor Infiltrating Lymphocytes combined score; SABR, stereotactic ablative body radiotherapy.

When applying quantitative habitat analysis to determine makeup of the whole breasts on the pretreatment MRIs, the largest %HM was L4 (28%), followed by L3 (22%). The smallest was H1 (4%), followed by H2 (5%). On a whole breasts basis, there was no significant change in individual or combination habitat composition between pre- and post-SABR breast MRIs, with changes of 2% or less for each habitat (Figure 2).

Figure 2.

Pie chart representation of tumor and whole breast % makeup for individual habitats pre- (A and B) and post-SABR treatment (C and D). Individual habitats are represented by corresponding colors and numbers in the legend (red 1 = H1, orange 2 = H2, green 3 = H3, yellow 4 = H4, violet 5 = L1, lavender 6 = L2, blue 7 = L3, brown 8 = L4). Abbreviation: SABR, stereotactic ablative body radiotherapy.

For quantitative habitat makeup analysis of the individual tumors, habitat H4 (high enhancing, latest phase) was the most represented on pre-SABR MRIs, making up an average of 37% of the tumors, followed by habitat H3 (22%). Tumor makeup was significantly different than whole breast habitat makeup. The largest pre-SABR difference between individual habitats in the tumor versus the whole breasts was seen with habitat H4 representing 37% of the tumor versus only 9% of the whole breast (P < 0.00001, Figure 2). The difference was even more pronounced on the post-SABR MRIs, where habitat H4 increased representation to 50% of tumor makeup compared to 9% of the whole breast (P < 0.00001, Figure 2). For habitat combinations, the biggest difference between tumor and whole breasts was seen with the high-intensity habitats (H1–4) making up 84% of tumors versus only 25% of the whole breasts on pre-SABR MRIs. Pre-SABR, the differences between individual habitat makeups in the tumor versus the whole breasts were statistically significant for all habitats except H1, with P-values of the remaining habitats ranging 0.005 to <0.00001. Post-SABR, the differences between tumor habitats and whole breast habitats were statistically significant for habitats H3 and H4 (P = 0.03 and <0.00001, respectively). Thus, after SABR treatment, representation of habitats H1 and H2 in the tumors was not significantly different than background whole breasts.

Comparing tumor habitats pre- and post-SABR, late phase habitat H4 showed the largest change by increasing 13% (P = 0.039). Conversely, habitats H1, H2, and H3 demonstrated individual decreases of 3%–8% and 17% overall decrease (Table 2 and Table 3). Examples of tumor habitat combination segmentation color overlay for habitats H1–H3 versus remaining habitats combined are shown in Figure 3 and Figure 4.

Table 2.

Habitat Tumor Composition on Pre- and Post-SABR MRI

Habitat	% of Tumor Pretreatment (Average)	% of Tumor Post-treatment (Average)	% Change (Average)	P-value
H1	11.1%	7.7%	−3.4%	0.560
H2	14.2%	6.5%	−7.7%	0.042a
H3	21.6%	15.5%	−6.1%	0.192
H4	36.7%	49.6%	+ 12.9%	0.039a
L1	1.3%	1.6%	+0.3%	0.764
L2	3.1%	1.6%	−1.5%	0.463
L3	4.6%	4.4%	−0.2%	0.911
L4	7.5%	13.1%	+5.6%	0.236

Percentages in the table do not add exactly to 100 because of rounding.

Abbreviation: SABR, stereotactic ablative body radiotherapy.

^aDenotes significance of P < 0.05 on paired t-test.

Table 3.

Percent of Total Tumor Representation for Habitat Combinations Averaged on the Pre treatment MRI and on the Post-treatment MRI

Habitat Combinations	% of Tumor Pretreatment (Average)	% of Tumor Post-treatment (Average)	% Change (Average)	P-value
H1–4	83.6%	79.3%	−4.3%	0.542
H1–3	46.8%	29.7%	−17.1%	0.006a
H1–2	25.2%	14.2%	−11%	0.079
H3–4	58.3%	65.1%	+6.8%	0.378

Differences in percentages between the pre- and post-treatment MRIs were analyzed using paired t-test.

^aDenotes significance of P < 0.05.

Figure 3.

64-year-old patient with right breast invasive ductal carcinoma at 10 o’clock with pathologic partial response after SABR. Single slice post-contrast axial MRI shows the right breast with segmentation habitat color overlay (arrows) before (A and zoomed B) and after (C and zoomed D) SABR treatment. Individual voxels within the segmented tumor are assigned red (habitats H1–3) or yellow (habitats H4 and L1–L4) color overlay. This patient had 20% tumor cellularity on pathology and 7% volume remaining of habitats H1–3. Abbreviation: SABR, stereotactic ablative body radiotherapy.

Figure 4.

65-year-old patient with right breast 6-mm invasive ductal carcinoma at 9 o’clock with pathologic nonresponse. Single slice post-contrast axial MRI shows the right breast with segmentation habitat color overlay (arrows) before (A and zoomed B) and after (C and zoomed D) SABR treatment. Individual voxels within the segmented tumor are assigned red (habitats H1–3) or yellow (habitats H4 and L1–4) color overlay. This patient had 80% tumor cellularity on pathology and 92% volume remaining of habitats H1–3. Abbreviation: SABR, stereotactic ablative body radiotherapy.

A decrease in %HM of H1–3 combined, when used as a diagnostic test to predict pPR versus pNR, demonstrates an accuracy of 94% (93% sensitivity, 100% specificity, 100% PPV, and 67% NPV). Compare this to a change in either lesion type or kinetic descriptors in the MRI report, which predicted pPR with 74% accuracy (sensitivity 71%, specificity 100%, PPV 100%, NPV 29%) (12). A decrease in H1–3 %HM is also associated with lower tumor cellularity on average compared to an increase after treatment (32% vs 60%, respectively, P = 0.02).

Analysis of %VR using various methods had strong linear correlation with %TC (R = 0.7023–0.8932) (Table 4). The %VR by 3D cubic volume from the MRI report was strongly correlative (R = 0.7709) (Figure 5). When quantitative microhabitat %VR was evaluated, some habitat combinations had even higher linear correlations with %TC. First, total tumor segmentation %VR was evaluated (all eight habitats combined), and linear correlation with %TC was strong (R = 0.7023) but lower than reported 3D %VR. Eliminating the subthreshold L1–4 habitats from the analysis resulted in improved linear correlation for combined habitats H1–4 (R = 0.8044). Since habitat H4 increased representation on average in the tumor after SABR treatment while habitats H1–3 decreased representation, H4 was removed, and higher correlation R was observed for %VR of H1–3 (R = 0.8932) (Figure 6).

Table 4.

Pearson’s Correlation Coefficient Values for Various Treatment Response %VR Measurements versus %TC at Final Pathology, Ordered by Correlation Strength as Determined by R Value

%VR Analysis	R value	P-value
Habitats H1–3	0.8932	<0.00001a
Habitats H1–4	0.8044	0.0002a
Reader volume	0.7709	0.0005a
Habitats H1 + H2	0.8244	0.00009a
All habitats (H1 through L4)	0.7023	0.002a
Habitats H3 + H4	0.1617	0.55

Abbreviations: %TC, percent tumor cellularity; %VR, percent volume remaining.

^aDenotes significant value for P < 0.05.

Figure 5.

Scatter plot displaying percent tumor cellularity (%TC, x-axis) and post-SABR percent volume remaining (%VR, y-axis) for each tumor. Line demonstrates a linear best fit. Abbreviation: SABR, stereotactic ablative body radiotherapy.

Figure 6.

Scatter plot displaying percent tumor cellularity (%TC, x-axis) and post-SABR percent volume remaining (%VR) for habitats H1–3 (y-axis) for each tumor. Line demonstrates a linear best fit. Abbreviation: SABR, stereotactic ablative body radiotherapy.

To evaluate immune response in the tumor, %HM and %VR were correlated with TILs in the pathology specimens, which ranged from 5% to 20% (mean 11%, standard deviation 5%). For habitat comparisons, habitat H4 makeup in the post-treatment MRI demonstrated moderate linear correlation with TILs (R = 0.5296, P = 0.035) despite not showing correlation on the pretreatment scans (R = 0.1507, P = 0.577). No other habitat had statistically significant correlation.

To evaluate the combined %TC and TIL value, CelTIL scores were calculated. CelTIL scores ranged from −57.5 to −3 and were scaled 0–100. Correlating with habitat changes from pre- to post-treatment MRIs, the change in habitat makeup for habitats H1 and H4 showed the highest, though moderate, linear correlations with CelTIL scores (R = −0.5399 and R = 0.4206, respectively). In evaluating habitat combinations, change in habitats H1–3 had the highest, moderate linear correlation with CelTIL (R = −0.5964). Thus, increases in habitat H4 and decreases in habitats H1–3 correlated best with the combined %TC and TIL score.

Evaluation of %VR compared to CelTIL demonstrated similar correlations compared to %TC alone. CelTIL scores had the strongest negative linear correlation with the combined %VR of habitats H1–3 (R = −0.9037, P < 0.00001). This was higher than the correlation with reported 3D %VR (R = −0.7784), %VR of habitats H1–4 (R = −0.7454), and %VR of all habitats combined (R = −0.6637).

Discussion

The results of this pilot study evaluating habitats within ER/PR+ HER2− tumor response to preoperative SABR demonstrate potential advantages over standard reported volume and kinetic curve results. Indeed, a decrease in %HM of H1–3 showed higher accuracy for prediction of pPR than change in reported lesion descriptors; similarly predicted lower %TC and %VR of H1–3 demonstrated the highest correlation with %TC. In the future, this could be used to adjust treatment in patients defined as nonresponders on MRI and could also potentially lead to dose or treatment adjustments aimed at achieving pCR in partial responders. It may be that the treatment dose in this phase II trial was not high enough or not given over a long enough period to achieve pCR. In a preoperative radiation treatment feasibility study by Wang and colleagues, dose response was demonstrated as MRI parameters showed linear correlation with treatment dose. Perhaps the patients who responded best in our study (tumor cellularity as low as 20% at surgical pathology) would continue to respond with a higher dose or dose given over a longer period. In the dose and time to imaging breast cancer SABR trial by Mouawad and colleagues, the cohort with the longest time to imaging (2.5 weeks post-SABR) and similar dose to our study (30 Gy given over three fractions) was deemed the best cohort for MRI evaluation as the longer time before imaging after treatment reduced the confounding acute effects of the radiation treatment. These six patients all showed decrease in percent change for signal enhancement area under the curve, ranging from 14% to 87% decrease. These results are similar to our %VR analyses. However, our 5–6 weeks post-SABR MRI imaging is the longest of the breast SABR MRI studies to date. This contrasts with many neoadjuvant chemotherapy studies in which response is evaluated over the course of months (24–26).

The fact that our various methods of evaluating tumor volume response on MRI correlated well with pathologic response lends clinical promise. For instance, studies have shown tumor cellularity correlates with volume reduction on MRI for NAC (27,28), and tumor cellularity as a factor in residual cancer burden calculation has shown benefit in identifying relapse risk (29,30). Thus, even in patients who do not achieve pCR, there is clinical benefit for lower residual cancer burden. Therefore, quantitative analysis of MRI response not only has the potential to guide treatment choices but also to predict future relapse risk. Indeed, where Jafri and colleagues showed utility in analyzing functional tumor volume (FTV) to predict recurrence-free survival, analysis of certain habitat combinations in our study is similar to their analysis of functional tumor volume (31). Where they defined functional tumor volume as summation of voxels meeting percent enhancement threshold of 70% and signal enhancement threshold of 1.0, a similar evaluation of habitat combinations in our study could be thought of as viable tumor volume (VTV). For instance, since the combinations of habitats H1–3 most closely correlated with %TC at pathology, future studies could evaluate the utility of analyzing this combination as a VTV biomarker.

In addition to prediction of tumor response to neoadjuvant radiation treatment, the quantitative analysis performed in this study has the potential to identify malignant tumors from background breast tissue. Adapting a technique used at our institution in studies of glioblastoma multiforme tumors in the brain, this was the first time this analysis was performed on breast tumors (16). We demonstrated that the habitats in tumors and whole breasts are distinctly different. This comparison of tumor habitats to whole breast habitats allows unique opportunities for evaluation of tumors at baseline as well as in treatment response monitoring. In a meta-analysis of 44 studies evaluating breast cancer NAC on MRI, Marinovich and colleagues found that specificity for residual tumor was greater when negative MRI was defined as contrast enhancement less than or equivalent to normal background parenchymal tissue enhancement compared to no enhancement (2). Indeed, our study demonstrated that many of the changes in the habitat of the tumors responding to treatment were changes making them more similar to background whole breast habitats through reductions in habitats H1–3 and increases in subthreshold habitats. The biggest exception was the tumor increases in habitat H4 makeup, and this particular habitat may reflect a unique biology in the tumors responding to treatment compared to background. Previous studies have shown that radiation therapy has effects on tumor vasculature, causing endothelial cell apoptosis and senescence (32). A study of radiation doses of 15–30 Gy given to tumors in mice demonstrated vascular damage and accompanying tumor microenvironment deterioration, for example (33). As habitat H4 is the latest phase high-enhancing habitat, one may speculate that this habitat reflects tumor enhancement in a more disrupted vascular environment within the tumor, leading to later high enhancement. Another possibility is that tumor-infiltrating lymphocytes are contributing to the change in habitat and are represented by an increase in a particular habitat, such as habitat H4, or habitat combination. This possibility is supported by the fact that post-SABR habitat H4 makeup showed the highest linear correlation of the individual habitats with TILs. Future studies should include TILs evaluations to elucidate the role habitat H4 might play not only as a tumor response biomarker but also as a tumor immune response biomarker.

This pilot study has several limitations. First, this was the first time this type of quantitative analysis was applied to breast tumors, thus limiting comparison to other neoadjuvant studies. Additionally, as a phase II trial, sample size was small and limits power. However, the strong correlations of tumor response on MRI to pathologic response suggest these results should be further evaluated in larger studies. Next, the selection of patients for this phase II trial was for ER+ HER2− breast cancer patients with tumors <2 cm in size. We know from NAC studies that these tumors typically respond less robustly to neoadjuvant treatment than more aggressive breast cancer phenotypes (34). Thus, quantitative analysis should be expanded to those phenotypes in future studies to determine if the quantitative response correlations demonstrated in this study are also applicable to additional breast cancer phenotypes. Finally, quantitative analysis was limited by the need for user input in tumor segmentation, lending some operator dependence to the quantitative analysis. Future studies could incorporate multiple readers for segmentation or segmentation agnostic approaches.

Conclusion

Quantitative habitat analyses of %HM and %VR in patients treated with SABR demonstrated higher accuracy for predicting pathologic response to therapy than standard reported lesion volume and descriptor analyses. In addition, microhabitat %VR analysis was similar to NAC studies evaluating FTV in predicting %TC, while %HM analysis demonstrating a shift to later phase enhancement similar to background was also consistent NAC response studies. This type of analysis has the potential to guide treatment prior to surgery, following future validating studies.

Key Messages.

• Quantitative habitat analyses of percent habitat makeup (%HM) and percent volume remaining (%VR) in patients with early-stage ER/PR+ HER2− breast cancer treated with preoperative stereotactic ablative body radiotherapy (SABR) yielded results predictive of pathologic response.

• This analysis demonstrated increased utility over standard reported MRI tumor volume response.

• Microhabitat %VR analysis was similar to neoadjuvant chemotherapy (NAC) studies evaluating functional tumor volume in predicting percent tumor cellularity.

• %HM analysis demonstrating a shift to later phase enhancement closer to background parenchymal enhancement was also consistent with NAC response studies.