Dynamic Diffuse Optical Tomography for Monitoring Neoadjuvant Chemotherapy in Patients with Breast Cancer

Abstract

Purpose

To identify dynamic optical imaging features that associate with the degree of pathologic response in patients with breast cancer during neoadjuvant chemotherapy (NAC).

Materials and Methods

Of 40 patients with breast cancer who participated in a longitudinal study between June 2011 and March 2016, 34 completed the study. There were 13 patients who obtained a pathologic complete response (pCR) and 21 patients who did not obtain a pCR. Imaging data from six subjects were excluded from the study because either the patients dropped out of the study before it was finished or there was an instrumentation malfunction. Two weeks into the treatment regimen, three-dimensional images of both breasts during a breath hold were acquired by using dynamic diffuse optical tomography. Features from the breath-hold traces were used to distinguish between response groups. Receiver operating characteristic (ROC) curves and sensitivity analysis were used to determine the degree of association with 5-month treatment outcome.

Results

An ROC curve analysis showed that this method could identify patients with a pCR with a positive predictive value of 70.6% (12 of 17), a negative predictive value of 94.1% (16 of 17), a sensitivity of 92.3% (12 of 13), a specificity of 76.2% (16 of 21), and an area under the ROC curve of 0.85.

Conclusion

Several dynamic optical imaging features obtained within 2 weeks of NAC initiation were identified that showed statistically significant differences between patients with pCR and patients without pCR as determined 5 months after treatment initiation. If confirmed in a larger cohort prospective study, these dynamic imaging features may be used to predict treatment outcome as early as 2 weeks after treatment initiation.

^© RSNA, 2018

Online supplemental material is available for this article.

Introduction

Breast cancer affects one in eight women in her lifetime. In the United States, there are approximately 230 000 women with invasive breast cancer, and about 40 000 die every year (1). Neoadjuvant chemotherapy (NAC) is widely applied as standard treatment for patients with newly diagnosed operable invasive breast cancer (2). The goal of NAC is to shrink the breast tumor prior to surgical intervention. As a result of NAC, an inoperable cancer may become operable, or surgery for an operable cancer can be converted from mastectomy to breast-conserving therapy. In addition, one of the advantages of NAC is that the responsiveness of a patient’s breast tumor to systematic therapy can be assessed prepoperatively. Pathologic response to NAC also associates with breast cancer recurrence and survival. Ideally a pathologic complete response (pCR), meaning the complete pathologic disappearance of the tumor before surgery, is achieved. Studies have shown that patients with a pCR have a 90%–95% 5-year survival rate, while 5-year survival rates for patients without pCR are typically less than 80% (3). Depending on the exact definition, pCR is achieved in only 15%–40% of women. Recent studies have shown that patients who do not show an early response may benefit from a change in treatment regimen (4–9). Therefore, it would be highly desirable to reliably determine who will and will not respond to NAC early in the treatment (10,11).

Over the past decade, diffuse optical tomography (DOT) has emerged as an imaging modality that may be used in breast cancer imaging and is potentially capable of early prediction of treatment outcome (11–13). DOT uses optical transmission measurements with nonionizing, low-intensity near-infrared light to generate three-dimensional maps of various tissue parameters. Optical property contrast from endogenous chromophores (oxyhemoglobin [HbO₂], deoxyhemoglobin [Hb], water, and lipid) can distinguish malignant from normal tissue (14–17). For example, it has been reported that total hemoglobin levels, which relate to tumor blood vessel density, are twice as high in malignant tumors as in benign breast lesions (18). Furthermore, in small-animal models of tumors, it has been reported that vascular changes precede measurable structure changes (19).

However, previous optical imaging studies have included patients receiving multiple different heterogeneous NAC regimens (20–23,27,29–34). Some studies even included chemotherapy and radiation therapy (32,34). The use of different treatment regimens for patients in the same cohort may confound some of the obervations. For example, different drugs have differing antiangiogenic effects and may affect optical signals in disparate ways. Recently, Schaafsma et al (35) reported results in 22 patients who were all treated with the same NAC protocol. All patients received six cycles of docetaxel, doxorubicin, and cyclophosphamide (TAC) with or without zoledronic acid. Using a non-tomographic diffuse optical spectroscopy system, they investigated the tumor-bearing breast. They found a statistically significant difference in the decrease in HbO₂ and Hb concentrations after the first cycle of NAC between patients with a pCR and those with a partial response.

Our group has previously used this dynamic diffuse optical tomography (DDOT) system to observe the hemodynamics of different tumors to determine if a subject had a benign or malignant tumor or no tumor at all. The contrast mechanism that was used for that diagnostic study was a simple breath hold. The quantified hemodynamic time traces showed variation between the healthy and tumor-bearing breasts of patients compared with those of healthy subjects, which showed similar traces for both breast. Additionally, we observed a more pronounced change in deoxyhemoglobin (Δ[Hb]%) in the tumor at the 15-second post–breath hold (15).

Given these promising results, we hypothesized that we will be able to observe the vascular changes that occur in tumors by using DDOT early in treatment. The purpose of the study was to identify optical biomarkers that can be used to predict pathologic response in patients with breast cancer during NAC.

Materials and Methods

Study Subjects and Procedures

This study was compliant with the Health Insurance Portability and Accountability Act and was approved by the institutional review board. Each subject gave written informed consent to the study. Between June 2011 and March 2016, we recruited 40 women (mean age, 42.2 years ± 8.2) over the age of 18 who had been given a diagnosis of stage II or stage III breast cancer in a study in which patients were enrolled prospectively (therefore, we did not know whether they would respond to treatment). The diagnosis was determined with x-ray mammography and biopsy. The primary tumor had to be at least 1 cm in diameter. Women with metastases were not eligible. All women in the study received the same treatment: 12 weekly cycles of a taxane (paclitaxel, docetaxel, or abraxane) followed by four cycles of doxorubicin and cyclophosphamide every 2 weeks with growth-factor support. At the end of the treatment, breast cancer surgery was performed, and any remaining tumor mass was removed and evaluated at histopathologic examination. For each person, the residual cancer burden (RCB) score G_RCB was calculated (36). A person was considered to have a pCR if the G_RCB ≤ 0.5 (RCB class 0). The RCB score was further classified as RCB-I (0.5 < G_RCB ≤ 1.36), RCB-II (1.36 < G_RCB ≤ 3.28), and RCB-III (G_RCB > 3.28) (38). Patients identified as belonging to the RCB-III group were the least responsive to NAC.

DDOT Instrumentation

DDOT measurements were performed in all patients just before treatment started and 2 weeks later. The machine used for this study was a continuous-wave dynamic imaging system that provides full three-dimentional tomographic images of both breasts at a rate faster than 1 second. A more detailed description of the system can be found in Appendix E1 (online) and in previous reports (14,37,38). Imaging was performed by researchers (J.E.G. and M.F., with 5 years of experience in optical imaging systems or M.A., with 2 years of experience in optical imaging), an oncologist (E.A.L., with 3 years of experience), or a research coordinator (J.C., with 2 years of experience).

Imaging Procedure

After a patient’s breasts were placed in the imaging probe, we asked her to remain as still as possible during the image acquisition. Imaging began with a baseline measurement of about 2 minutes. Next, we imaged the patient while she was holding her breath. We asked patients to hold their breath for up to 30 seconds. All patients were able to hold their breath for at least 15 seconds, and all data were included in our analysis. Imaging continued for another 90 seconds after the patient concluded her breath hold and started breathing again. Each patient performed two or three breath-hold cycles. The total time for the entire imaging procedure was 5–10 minutes (Fig 1).

Figure 1:

Timeline for subject imaging. At the beginning, the patient remains stationary for the first 2 minutes. Subsequently, subjects perform three breath holds (BH) for about 30 seconds and are given 90 seconds to recover. Data selection for dynamic reconstructions is performed 30 seconds before one of the breath holds for the baseline. Each frame of the breath hold and 30 seconds of recovery time are reconstructed to find the percentage change in oxy- and deoxyhemoglobin from the breath-hold baseline.

After the data acquisition was completed, data from one of the breath holds was selected for three-dimensional image reconstruction by one of the researchers (J.E.G., M.A.) for analysis. The second breath hold was usually chosen, unless substantial patient motion or coughing was observed, in which case the third breath hold was used. Approximately 2–3 images per second were reconstructed (depending on the number of sources used in imaging, the frame rate was 1.74 frames/sec [32 sources] to 3.09 frames/sec [18 sources]). Averaged data from 30 seconds before the breath hold were used as a baseline against which changes in deoxy- and oxyhemoglobin (S_Hb(t) = %Δ[Hb](t) and S_HbO2(t) = %Δ[HbO₂](t)) were calculated. Data from 30 seconds after the conclusion of the breath hold were used to obtain the recovery images (Fig 1). Figures 2 and 3 show examples of images obtained in patients with breast cancer.

Figure 2a:

Images in a 46-year-old postmenopausal woman with a poorly differentiated invasive ductal carcinoma with a pathologic complete response to neoadjuvant chemotherapy (NAC) who was assigned to RCB class 0 (RCB-0). Her baseline tumor had an initial maximum diameter of 2.9 cm as determined at mammography (a clinical palpation examination found the diameter to be 5.5 cm). Image a displays an axial mammogram obtained prior to chemotherapy. Images in b are sagittal diffuse optical tomography images obtained in the left and right breasts just before therapy. The images displayed refer to a time point of 15 seconds after the breath hold and show the percentage change in deoxyhemoglobin (%Δ[Hb]). Finally, image c shows time-dependent signal traces in response to a 30-second breath hold for the tumor region obtained at dynamic diffuse optical tomography just before the start of NAC (baseline) and 2 weeks after treatment initiation.

Figure 3a:

Images in a 61-year-old postmenopausal woman with a moderately differentiated invasive lobular carcinoma show a patient with no response (RCB-III) to neoadjuvant chemotherapy (NAC). With MR imaging, it was determined that she had a large tumor of 9.5 cm in diameter at the start of the therapy. She did not respond well to NAC and was classified as RCB-III at the time of surgery. Image a displays an axial MR image obtained prior to chemotherapy. Image b are sagittal diffuse optical tomography images obtained from the left and right breasts just before therapy. The images displayed refer to a time point of 15 seconds after the breath hold and show the percentage change in deoxyhemoglobin (%Δ[Hb]). Finally, image c shows time-dependent signal traces in response to a 30-second breath hold for the tumor region obtained from dynamic diffuse optical tomography just before the start of NAC (baseline) and 2 weeks after treatment initiation.

Figure 2b:

Images in a 46-year-old postmenopausal woman with a poorly differentiated invasive ductal carcinoma with a pathologic complete response to neoadjuvant chemotherapy (NAC) who was assigned to RCB class 0 (RCB-0). Her baseline tumor had an initial maximum diameter of 2.9 cm as determined at mammography (a clinical palpation examination found the diameter to be 5.5 cm). Image a displays an axial mammogram obtained prior to chemotherapy. Images in b are sagittal diffuse optical tomography images obtained in the left and right breasts just before therapy. The images displayed refer to a time point of 15 seconds after the breath hold and show the percentage change in deoxyhemoglobin (%Δ[Hb]). Finally, image c shows time-dependent signal traces in response to a 30-second breath hold for the tumor region obtained at dynamic diffuse optical tomography just before the start of NAC (baseline) and 2 weeks after treatment initiation.

Figure 3b:

Images in a 61-year-old postmenopausal woman with a moderately differentiated invasive lobular carcinoma show a patient with no response (RCB-III) to neoadjuvant chemotherapy (NAC). With MR imaging, it was determined that she had a large tumor of 9.5 cm in diameter at the start of the therapy. She did not respond well to NAC and was classified as RCB-III at the time of surgery. Image a displays an axial MR image obtained prior to chemotherapy. Image b are sagittal diffuse optical tomography images obtained from the left and right breasts just before therapy. The images displayed refer to a time point of 15 seconds after the breath hold and show the percentage change in deoxyhemoglobin (%Δ[Hb]). Finally, image c shows time-dependent signal traces in response to a 30-second breath hold for the tumor region obtained from dynamic diffuse optical tomography just before the start of NAC (baseline) and 2 weeks after treatment initiation.

To capture the features of the time-dependent response curves, S_Hb(t) and S_HbO2(t) (see Figs 2c and 3c) in a more quantitative way, we used the following nine parameters, many of which have been used in other dynamic imaging modalities and previous studies (15,39): initial enhancement (IE) of the breath hold, post-initial enhancement (PIE), rise slope (m_rise), fall slope (m_fall), rise rate (q_rise), washout rate (q_fall), the peak value of the tumor trace (S_peak), and the normalized maximum peak value (NMPV). As the ninth feature, we calculated the correlation coefficient (CC) between the time traces observed in the healthy and tumor-bearing breasts in corresponding volume of interest. (See Appendix E1 [online] for a detailed explanation of how these parameters were extracted from S_Hb(t) and S_HbO2(t)).

Figure 2c:

Images in a 46-year-old postmenopausal woman with a poorly differentiated invasive ductal carcinoma with a pathologic complete response to neoadjuvant chemotherapy (NAC) who was assigned to RCB class 0 (RCB-0). Her baseline tumor had an initial maximum diameter of 2.9 cm as determined at mammography (a clinical palpation examination found the diameter to be 5.5 cm). Image a displays an axial mammogram obtained prior to chemotherapy. Images in b are sagittal diffuse optical tomography images obtained in the left and right breasts just before therapy. The images displayed refer to a time point of 15 seconds after the breath hold and show the percentage change in deoxyhemoglobin (%Δ[Hb]). Finally, image c shows time-dependent signal traces in response to a 30-second breath hold for the tumor region obtained at dynamic diffuse optical tomography just before the start of NAC (baseline) and 2 weeks after treatment initiation.

Figure 3c:

Images in a 61-year-old postmenopausal woman with a moderately differentiated invasive lobular carcinoma show a patient with no response (RCB-III) to neoadjuvant chemotherapy (NAC). With MR imaging, it was determined that she had a large tumor of 9.5 cm in diameter at the start of the therapy. She did not respond well to NAC and was classified as RCB-III at the time of surgery. Image a displays an axial MR image obtained prior to chemotherapy. Image b are sagittal diffuse optical tomography images obtained from the left and right breasts just before therapy. The images displayed refer to a time point of 15 seconds after the breath hold and show the percentage change in deoxyhemoglobin (%Δ[Hb]). Finally, image c shows time-dependent signal traces in response to a 30-second breath hold for the tumor region obtained from dynamic diffuse optical tomography just before the start of NAC (baseline) and 2 weeks after treatment initiation.

Statistical Analysis

For the statistical analysis of the data ,we used the IBM SPSS Statistics software package. Given the larger number of features considered (nine parameters that describe S_Hb(t) and S_HbO2(t)), we first performed a Pearson correlation to determine if there were linear dependencies on any pair of parameters (Appendix E1 [online]). We found particularly strong correlations (|P_corr| > .75) between all Hb and HbO₂ pairs, meaning that the nine features extracted from the S_Hb(t) traces and the related nine features extracted from S_HbO2(t) traces showed very strong correlations. Hence, going forward, we considered only the nine features deducted from S_Hb(t) traces. Other strong correlations (|P_corr| > .5, with significance at the .01 level) were found between m_rise and S_peak, q_fall and PIE, m_fall and m_rise and m_fall and PIE, and q_rise and NMPV. Weaker correlations (.5 > |P_corr| > .35, with significance at the .05 level) were determined for m_rise and IE, m_fall and q_fall, q_fall and IE, and PIE and S_peak.

Considering feature pairs that were significantly correlated (P_corc > .5), left only five features (out of the total of 18) that could be considered independent. Applying a Bonferroni correction for multiple comparison resulted in a 95% confidence interval for P < .01 (= .05/5); 90% confidence intervals were reached for P < .02 (= .1/5).

Means and standard deviations were determined for each of the nine time-trace parameters. One-way analyses of variance were performed for comparison of more than two groups. Otherwise, two-way unpaired student t tests between designated groups were performed. Receiver operating characteristic (ROC) curves were developed for specified features. The sensitivity, specificity, area under the ROC curve, accuracy, positive predictive value, and negative predictive value were determined from the Youden index of the ROC curve. Sensitivity in this article denotes the percentage of the more responsive group that was diagnosed correctly. Conversely, the specificity denotes the percentage of subjects in the least responsive group that were categorized correctly. The positive predictive value is the percentage of subjects who were classified in the responsive group that were truly responsive subjects. The negative predictive value is the percentage of the subjects who were classified in the least responsive group who truly were in the least responsive group.

Results

Patient Summary

Of the 40 patients enrolled in the study, six patients are not included in the analysis presented here. One patient completely stopped NAC midway through the therapy and was removed from the study. Two subjects missed their week-2 DDOT imaging time point. For another two subjects, instrumentation failure during the imaging session resulted in corrupted data sets that could not be used. Finally, one subject opted to receive radiation therapy in addition to traditional NAC therapy and therefore was removed from the analysis.

The average age of the remaining 34 patients was 49 years. There were 19 premenopausal women and 15 postmenopausal women. Four women had a diagnosis of invasive lobular carcinoma, and 29 patients had invasive ductal carcinoma. One subject had mixed invasive lobular and ductal carcinoma. Using the largest measurable dimension as the size metric, the average initial tumor size was 4.6 cm ± 2.8 based on baseline clinical examinations. After NAC, 13 patients were classified as RCB-0 (ie, pCR). Two patients were classified as RCB-I, 13 as RCB-II, and six as RCB-III. Table E1 (online) provides a full summary of the clinical and pathologic features.

Statistical Analysis of Hemodynamic Parameters

Analysis of hemoglobin traces observed at week 2—analysis of variance and P values.—Table 1 shows the data for all four cohorts (RCB-0, I, II, and III) for all nine features (IE, PIE, m_rise, m_fall, q_rise, q_fall, S_peak, NMPV, and CC) for the S_Hb(t) response curves 2 weeks after treatment initiation. The m_rise, S_peak, NMPV, and CC did not show statistically significant differences between the response groups. The features that showed some significant differences (P < .01 for 95% confidence interval and P < .02 for 90% confidence intervals) between groups were IE (P < .013 for RCB 0, I, and II vs RCB III), PIE (P < .004 for RCB 0 vs RCB I, II, and III and P < .007 for RCB 0 and I vs RCB II and III), m_fall (P < .006 for RCB 0 vs RCB I, II, and III and P < .002 for RCB 0 and I vs RCB II and III), and q_fall (P < .003 for RCB 0 vs RCB I, II, and III and P < .015 for RCB 0, I vs RCB II and III).

Table 1.

Summary of Mean Values for All Four RCB Classes for Features Derived from Time Traces of the Hemodynamic Responses, S_Hb(t) (Percentage Change in Deoxyhemoglobin Signal), Obtained 2 Weeks after the Initial Treatment with Taxane

Note.—Unless otherwise specified, data are means ± standard deviations. ANOVA = analysis of variance, CC = correlation coefficient, IE = initial enhancement, m_fall = fall slope, m_rise = rise slope, NMPV = normalized maximum peak value, pCR = pathologic complete response, PIE = postinitial enhancement, q_fall = washout rate, q_rise = rise rate, RCB = residual cancer burden, S_Peak = peak signal. P < .01 indicates a significant difference (95% confidence intervals–90% confidence intervals are reached for P < .02).

Sensitivity analysis for selected week-2 features.—Because the difference in q_fall for S_Hb traces between patients with a pCR (RCB 0) and those without a pCR (RCB I, II, and III) produced the strongest significant distinction (P < .003), we created a receiver operating characteristic (ROC) curve for this variable. The sensitivity, specificity, and accuracy were 92.3% (12 of 13 subjects), 76% (16 of 21 subjects), and 82.4% (28 of 34 subjects), respectively (cutoff point for q_fall = 0.09). The area under the ROC curve was 0.85, and the positive predictive value was 70.6% (12 of 17 subjects), while the negative predictive value was 94.1% (16 of 21 subjects). Results of similar ROC analyses for m_fall, PIE, IE, and q_rise for S_Hb are shown in Table 2.

Table 2.

Summary of Binary Classification Test for Features That Showed the Most Significant Differences between Subject Groups Based on an Analysis of the Data Obtained at Week 2 from the Percentage Change in Deoxyhemoglobin Signal, S_Hb(t)

Note.—Please see the rightmost three columns of Table 1 for the most significant differences. Data in parentheses are raw data. AUC = area under the receiver operating characteristic curve, IE = initial enhancement, m_fall = fall slope, NPV = negative predictive value, pCR = pathologic complete response, PIE = postinitial enhancement, PPV = positive predictive value, q_fall = washout rate, q_rise = rise rate, RCB = residual cancer burden.

To specifically find patients with RCB III patients, the best features seem ed to be IE (P < .013) and q_rise (P < .027) for the S_Hb traces. The areas under the receiver operating characteristic curve for IE and q_rise were 0.71 and 0.83, respectively. The parameter IE yielded a high sensitivity of 96.4% (27 of 28) and a high positive predictive value of 90.0% (27 of 30), while the use of q_rise yielded a high specificity of 100% (six of six) and a high PPV of 100% (15 of 15).

Analysis of differences between baseline and week-2 imaging data.—In addition to evaluating the hemoglobin traces observed at week 2, we also analyzed differences between data obtained at baseline (week 0) and those obtained at week 2 (Table 3). In this table, the most notable entries are the low P values for Δm_fall determined from S_Hb traces. Of the six groupings considered, only RCB 0, I, and II versus RCB III showed a nonsignificant statistical difference (P < .64). For the clinically meaningful comparison of RCB 0 versus RCB I, II, and III, patients with pCR showed an increase in Δm_fall compared with patients without pCR subject (S_Hb: P < .0004). For Δm_fall for S_Hb we found that the sensitivity, specificity, and accuracy were 61.5% (eight of 13), 100% (21 of 21), and 85.3% (29 of 34), respectively, with an area under the receiver operating characteristic curve of 0.87 (Table 4). The positive predictive value was 100% (eight of eight subjects), and the negative predictive value was 80.8% (21 of 26 subjects). The 100% specificity signifies that all the patients without pCR were identified as having a non-pCR response. This is of particular importance, as these are exactly the patients who would benefit from a change in therapy.

Table 3.

Summary of Differences in Features Derived from Time Traces of the Hemodynamic Responses, S_Hb(t) (Percentage Change in Deoxyhemoglobin Signal), between Baseline and 2 Weeks after the Initial Treatment with Taxane

Note.—Unless otherwise specified, data are means ± standard deviations. ANOVA = analysis of variance, ΔCC = change in correlation coefficient, ΔIE = change in initial enhancement, Δm_fall = change in fall slope, Δm_rise = change in rise slope, ΔNMPV = change in normalized maximum peak value, ΔPIE = change in postinitial enhancement, Δq_fall = change in washout rate, Δq_rise = change in rise rate, ΔS_peak = change in peak signal, pCR = pathologic complete response, RCB = residual cancer burden.

Table 4.

Summary of Binary Classification Test for the Features That Showed the Most Significant Differences between Subject Groups Based on Features Derived from Time Traces of the Hemodynamic Responses, S_Hb(t), between Baseline and 2 Weeks after the Initial Treatment with Taxane

Note.—Please see the rightmost three columns of Table 3 for the most significant differences. Data in parentheses are raw data. AUC = area under the receiver operating characteristic curve, ΔIE = change in initial enhancement, Δm_Rise = change in rise slope, Δm_fall = change in fall slope, Δq_Rise = change in rise rate, Δq_fall = change in washout rate, NPV = negative predictive value, PPV = positive predictive value, RCB = residual cancer burden.

Discussion

We describe a relatively large cohort of patients who followed the same NAC regimen, a standard taxane/anthracycline-based therapy. Our optical imaging system performs three-dimensional DDOT, which provides information about hemodynamic effects in both breasts simultaneously. This allows a direct comparison of the tumor-bearing breast and the non–tumor-bearing breast in the same patient.

Our results showed that tumors in patients with pCR (RCB = 0) had higher washout rates, q_fall, by week 2 compared with tumors in patients without pCR (RCB I, II, or III). Using q_fall for S_Hb as a parameter to predict patients with pCR, we achieved a sensitivity of 92.3% (12 of 13 subjects) and a specificity of 76% (16 of 21 subjects). Additionally, Δm_fall of the S_Hb trace showed a significant difference between patients with pCR and those without. This may possibly indicate an increase in blood flow in the tumor. For this analysis, the highest area under the receiver operating characteristic curve (0.87) and specificity (100% [21 of 21 subjects]) were obtained.

Like many other chemotherapeutic drugs, taxanes disrupt the cell cycle to prevent proliferation and cause apoptosis. In addition, it has been shown that taxanes disrupt microvessels and decrease microvascular density (40). However, studies in mice have suggested that, instead of taxanes disrupting blood flow, the diameter of the tumor vessels and blood flow velocity inside the vessels actually increase when a tumor responds to taxanes. To explain this apparent paradox, it has been argued that solid tumors have a naturally high interstitial fluid pressure due to the compression of blood vessels by surrounding tumor cells, the disorganized tumor vascular network, and the absence of functional lymphatics (41–43). Chemotherapy leads to a reduction in tumor cell density, which in turn lowers interstitial fluid pressure, consequently increasing blood vessel diameter and blood flow velocity. These arguments are supported by another study (44) that demonstrated that interstitial fluid pressure decreased after patiets with breast cancer received the taxane paclitaxel.

We saw statistically significant increases in m_fall for patients with pCR between the baseline and week-2 imaging time points, suggesting that there may be increased blood flow in the tumor. Therefore, there is evidence in this study to suggest that the increased apoptosis and increased interstitial fluid pressure in tumors responding to taxanes may cause the increase in m_fall by the 2nd week of NAC and that these could be physiologically relevant parameters.

Our study had limitations. Some patients could not hold their breath for the entire desired length of 30 seconds. Not surprisingly, curve parameters that were not dependent on the breath-hold length, such as normalized peak values and washout rates, yielded more promising results than parameters that depended on breath holding. Furthermore, because a minimum tumor size of 1 cm in diameter was required for study participation, no statement can be made on how well this method works for smaller tumors. Also, other non–taxane-based treatment regimens will need to be tested, and the influence of different types of drugs on outcome predictability is a goal for future studies. We did not assess for repeatability of our measures, and we did not correlate with other imaging data to ensure that what we were measuring corresponded to tumor location, although prior studies from our group (14,15) have addressed these issues. Finally, this was a pilot study to identify the most promising imaging parameters. Hence, we used our data to determine thresholds for the various dynamic imaging features. This can lead to an overestimation of the results. A follow-up prospective study with a larger cohort that produces an independent data set is needed to confirm our findings.

In conclusion, we have developed an imaging approach to noninvasively extract information about the hemodynamic effects that occur in and around the tumors of patients with breast cancer. In this study, we demonstrated that certain features of the hemodynamic responses to a simple breath hold can be used to predict treatment outcome in patients undergoing NAC. Using q_fall for S_Hb as a parameter to predict patients with pCR, we achieved a sensitivity of 92.3% (12 of 13 patients) and a specificity of 76% (16 of 21 patients). Additionally, the difference in m_fall of S_Hb traces between the two imaging points showed significant differences between patients with pCR and those without pCR, where patients with pCR showed an increase. Finally, we determined that q_rise of the S_Hb trace is the best feature for comparing patients with RCB 0, I, and II and patients with RCB III (ie, the patients at highest risk for breast cancer). In this case, we were able to achieve 100% specificity (six of six subjects), which indicates that we could identify all nonresponders. These are patients who are least likely to respond to NAC and who hence may be prime candidates for a change of therapy.

Implication for Patient Care

• ■ If confirmed in a prospective study with a larger cohort, dynamic optical imaging parameters identified in this pilot study may be used for predicting treatment response in patients with breast cancer undergoing several months of neoadjuvant chemotherapy, as early as 2 weeks after treatment initiation.