Monitoring breast cancer response to neoadjuvant chemotherapy with ultrasound signal statistics and integrated backscatter

Background Neoadjuvant chemotherapy (NAC) is used in patients with breast cancer to reduce tumor focus, metastatic risk, and patient mortality. Monitoring NAC effects is necessary to capture resistant patients and stop or change treatment. The existing methods for evaluating NAC results have some limitations. The aim of this study was to assess the tumor response at an early stage, after the first doses of the NAC, based on the variability of the backscattered ultrasound energy, and backscatter statistics. The backscatter statistics has not previously been used to monitor NAC effects. Methods The B-mode ultrasound images and raw radio frequency data from breast tumors were obtained using an ultrasound scanner before chemotherapy and 1 week after each NAC cycle. The study included twenty-four malignant breast cancers diagnosed in sixteen patients and qualified for neoadjuvant treatment before surgery. The shape parameter of the homodyned K distribution and integrated backscatter, along with the tumor size in the longest dimension, were determined based on ultrasound data and used as markers for NAC response. Cancer tumors were assigned to responding and non-responding groups, according to histopathological evaluation, which was a reference in assessing the utility of markers. Statistical analysis was performed to rate the ability of markers to predict the final NAC response based on data obtained after subsequent therapeutic doses. Results Statistically significant differences (p<0.05) between groups were obtained after 2, 3, 4, and 5 doses of NAC for quantitative ultrasound markers and after 5 doses for the assessment based on maximum tumor dimension. Statistical analysis showed that, after the second and third NAC courses the classification based on integrated backscatter marker was characterized by an AUC of 0.69 and 0.82, respectively. The introduction of the second quantitative marker describing the statistical properties of scattering increased the corresponding AUC values to 0.82 and 0.91. Conclusions Quantitative ultrasound information can characterize the tumor's pathological response better and at an earlier stage of therapy than the assessment of the reduction of its dimensions. The introduction of statistical parameters of ultrasonic backscatter to monitor the effects of chemotherapy can increase the effectiveness of monitoring and contribute to a better personalization of NAC therapy.


Abstract
Neoadjuvant chemotherapy (NAC) is used in patients with breast cancer to reduce tumor focus, metastatic risk, and patient mortality. Monitoring NAC effects is necessary to capture resistant patients and stop or change treatment. The existing methods for evaluating NAC results have some limitations. The aim of this study was to assess the tumor response at an early stage, after the first doses of the NAC, based on the variability of the backscattered ultrasound energy, and backscatter statistics. The backscatter statistics has not previously been used to monitor NAC effects.
The B-mode ultrasound images and raw radio frequency data from breast tumors were obtained using an ultrasound scanner before chemotherapy and 1 week after each NAC cycle.
Twenty-four malignant breast cancers, qualified for neoadjuvant treatment before surgery, were included in the study. The shape parameter of the homodyned K distribution and integrated backscatter, along with the tumor size in the longest dimension, were determined based on ultrasound data and used as markers for NAC response. Cancer tumors were assigned to responding and non-responding groups, according to histopathological evaluation, which was a reference in assessing the utility of markers. Statistical analysis was performed to rate the ability of markers to predict the final NAC response based on data obtained after subsequent therapeutic doses. Statistically significant differences between groups were obtained after 2, 3, 4, and 5 doses of NAC for quantitative ultrasound markers and after 5 doses for the assessment based on maximum tumor dimension. After the second and third NAC courses the marker, which was a linear combination of both quantitative ultrasound parameters, was characterized by an AUC of 0.82 and 0.91, respectively.
The introduction of statistical parameters of ultrasonic backscatter to monitor the effects of chemotherapy can increase the effectiveness of monitoring and contribute to a better personalization of NAC therapy.

Introduction
Neoadjuvant chemotherapy (NAC) was initially used in locally advanced breast cancer (LABC) and in the case of inflammatory cancer [1]. Currently, it is also recommended in patients with triple-negative cancer (TNBC), with the presence of HER-2 + receptors (Luminal B HER2-positive and HER-positive non-luminal subtype), and in cases of luminal B HER2-negative tumors with low expression of hormone receptors, with high grade of malignancy (G3) in patients at an early age (≤ 35 years) in the second or third stage of cancer [2][3][4]. Neoadjuvant chemotherapy reduces the risk of metastases and micrometastases in distant organs. It also reduces the neoplastic focus and decreases the frequency of the recurrences and the mortality of patients.
The use of NAC therapy does not always bring the expected results. A meta-analysis conducted on a group of over 18,000 patients (from 49 studies) who received NAC showed that the pathologic complete response rate was 21.5% [5]. In the HER2-positive disease and the triple-negative disease groups, the percentages of complete responses for the treatment reached 76 and 67%, respectively [5]. The complete response is significantly lower in the case of the luminal B HER2-negative subtype, but patients with additional risk factors may also benefit from pre-operative treatment. According to available data, even over 40% of patients undergoing chemotherapy show a poor pathological response to the treatment [6]. In such cases, the whole cycle of neoadjuvant therapy, which lasts about 4 months (five cycles), is associated with the delayed start of another, more effective treatment and unnecessary exposure of the patient to the toxic effects of the drugs used. The response to the treatment varies and requires differentiation between responders and non-responders.
For monitoring, a clinical breast examination (CBE), mammography (MMG), traditional ultrasound imaging in B-mode (US), and magnetic resonance imaging (MRI) are used. Imaging with MRI is more accurate compared to CBE, US, or MMG; however, MRI is a technique with limited availability. Ultrasonography is considered a more accurate method in assessing tumor size and in the monitoring of residual breast tumors than CBE or MMG [7]. Recent work has also shown that classical US imaging techniques with sonoelastography are useful and allow predicting the response to the treatment with a high degree of sensitivity and specificity after the second course of chemotherapy. The decrease in tumor stiffness is a good predictor of a pathological response [8].
Methods based on monitoring changes in tumor size, which are determined on the basis of imaging tests or by palpation examination, have numerous limitations [7]. Changes in tumor size occur with a delay in relation to changes in the tumor microstructure. Sometimes, despite a positive pathological response to the treatment, there is no apparent reduction in primary tumor mass in imaging because it is masked by changes induced by the treatment.
Functional imaging techniques, such as positron emission tomography, MRI with contrast agents, diffusion weighted imaging, and diffuse optical spectroscopy, enable to capture changes in the microstructure, vascularization, and metabolic activity of tumors under the influence of chemotherapy after the first cycle of treatment [9][10][11]. However, these are costintensive and time-consuming methods which require intravenous administration of exogenous contrast agents to detect changes in the tumor after each course of chemotherapy.
Therefore, their use is limited.
The use of quantitative ultrasound (QUS) methods to assess the effectiveness of chemotherapy seems to be an interesting alternative to tracking changes in tumor size. The QUS techniques are based on the analysis of raw, ultrasonic radio-frequency echoes (RF) in order to determine the quantitative parameters characterizing the tissue. The QUS technique was also used to monitor the breast cancer response to chemotherapy. For example, Lin et al.
used animal models of breast cancer to show that the spectral analysis of ultrasonic echoes provides a way to assess the tumor response to chemotherapy [12]. Similarly, Sannachi et al.
carried out a study on a group of 30 tumors and showed that the use of a combination of quantitative measures: average scatterer diameter (ASD), and average acoustic concentration (AAC) allows for the differentiation of responding and non-responding patients, with a sensitivity of 82% and a specificity of 100% after the fourth week of treatment [13]. In later studies, Sannachi et al., based on a combination of quantitative parameters: texture, and molecular features, predicted tumor response to NAC with an accuracy of 79, 86, and 83% at weeks 1, 4, and 8 of the therapy, respectively [14]. Sadeghi-Naini et al. presented clinical trials using QUS on a group of 100 patients [15]; the authors used the mean values of quantitative parameters (mid-band fit, slope, intercept, spacing among scatterers, ASD, and AAC) as well as textural features of parametric maps based on these parameters. Their results show that QUS imaging is able to predict the outcome of chemotherapy in patients with breast

Patients
The study was carried out in the Department of Radiology, Maria Skłodowska-Curie Memorial Institute of Oncology in Warsaw, Poland.
The study included patients with breast cancer qualified for neoadjuvant chemotherapy (radial, radial + 45°, anti-radial, anti-radial +45°). The ultrasound examinations were performed in accordance with the American College of Radiology BI-RADS guidelines [16].
The region of interest (ROI) contoured around the tumor was determined by the same, experienced radiologist during each scan. The size of the tumor in the longest dimension was also measured.
The first scan was acquired before chemotherapy and was used as the baseline data. Subsequent ultrasound examinations were performed 1 week after each course of NAC; the patient was monitored for 5 to 6 months. All patients underwent the first four cycles of NAC, but some patients did not take part in subsequent stages of chemotherapy due to the oncologist's decision to perform a mastectomy or in the event of complete disappearance of the tumor.

Quantitative ultrasound parameters
Two QUS parameters were used, the integrated backscattering coefficient (IBSC) and the shape parameter of the homodyned K distribution. The hypothesis underlying their use is

Integrated backscatter coefficient
The integrated backscatter coefficient was derived from estimates of the backscatter coefficient (BSC), which is defined as the differential scattering cross section per unit solid angle at 180° per unit volume. The BSC was determined using the reference phantom method proposed by Yao et al. [19]. The technique allows limiting the impacts of system-dependent effects such as the system transfer function or diffraction artifacts by using data from the well-

Envelope statistics
The envelope statistics was based on the homodyned K distribution, which was first introduced by Jakeman [23]; this model of backscatter was chosen because of its flexibility.
The high usability of homodyned K distribution stems from the ability to describe statistics of envelopes of signals under varying conditions. This means that homodyned K distribution may be used when the number of scatterers is large and when they are uniformly distributed as well as when the number of scatterers is low or they are organized in periodical structures or in case of coherent component sources. Additionally, the shape parameter of the homodyned K distribution has shown that it may be used as an ultrasonic biomarker which characterizes the tissue microstructure [24]. The shape parameter ENS, also called "effective Additionally, a tumor size-based classifier was included in the analysis.
The statistical significance of the classifiers was assessed using p-values obtained with a twosided Wilcoxon rank sum test. The classifiers were cross-validated through the leave-one-out technique [27]. Evaluation of the results was based on the area under the receiver operating characteristic (ROC) curve (AUC) [28]; these values were determined for data acquired after each chemotherapy course. The results were analyzed for changes in classification effectiveness as a function of the therapy progress. All calculations were done using Matlab® 2018a (The MathWorks, Inc., Natick, MA).

Results and discussion
Pathologic responses to treatment Microscopic evaluation of the specimens revealed the presence of stroma-filled tissue (pink staining) with small isolated patches of glands (purple staining), demonstrating therapeutic effects (Fig 1a). In Fig 1b, high cellularity with low stromal collagen density is visible.
Assessment of the usefulness of QUS parameters and of the tumor size to predict the effects of chemotherapy was based on a comparison with RMC results.  The tumors were ranked according to the RMC value, which is marked on the vertical axis. On the horizontal axis, subsequent courses of chemotherapy were marked; the 24 circles correspond to 24 tumors. Those whose length has decreased by 30% or more are marked in white, while the tumors marked in black represent smaller changes or tumor growth. The 30% decrease was adopted as a limit value according to recommendations of the RECIST 1.1 (Response Evaluation Criteria in Solid Tumors [30]), dividing tumors into responders and non-responders depending on their maximum length change. Although ultrasonography is not recommended in the RECIST 1.1 guideline as a tool for assessing tumor changes during chemotherapy, these criteria are used in other scientific work [31].
In the group of tumors responding to NAC, according to histopathological assessment, there was a tendency to reduce the maximum tumor size along with subsequent doses of chemotherapy. After the end of the therapy, only in 3 out of 19 cases there were no changes to qualify the cancer, based on size reduction, as NAC-responsive tumors. In the group of N-PR tumors in one case, the assessment based on changes in size was inconsistent with an RMCbased assessment. The lack of size-based responses, when the response was positive, results from the fact that the invasion of fibrous stroma, caused by chemotherapy, results in the overestimation of tumor residue in the ultrasound [32].

QUS parametric maps
Cycles of representative B-mode ultrasound images for responding (0% RMC) and nonresponding (100% RMC) tumors are shown in Figure 3, together with the parametric overlays of the IBSC and ENS maps. They were calculated based on ultrasound data collected immediately before NAC treatment and 1 week after each NAC course. The pixel color reflects the parameter value and the position depends on the location of the sliding window.
Analysis of the IBSC map set revealed a clear upward trend in the IBSC values of responding tumors (Fig 3, first column). In the case of non-responding tumors, no such changes were observed (second column). For the responding tumor, the ENS values decreased as the treatment progressed (Fig 3, third column). In the case of a non-responding tumor, there was no clear trend in ENS changes (last column).

Fig 3. Examples of ultrasound B-mode images with IBSC and ENS parametric overlays.
Ultrasound data were obtained from a responding tumor (0% RMC) and non-responding tumor (100% RMC) prior to (first row) and 1 week after every NAC course (successive rows). The blue color indicates low values of both quantitative parameters, IBSC or ENS, while the red color indicates high values.

Responses of QUS parameters
Quantitative ultrasound parameters, mean IBSC and ENS values, were determined from the parametric maps. Changes in quantitative parameters during subsequent chemotherapy courses are shown in Figure 4 in relation to the parameter values before therapy. The form of the presentation is analogous to the presentation of the results of changes in maximum tumor size with one difference. Relative changes of parameters were divided into three ranges (presented in color-bar in Fig 4), reflecting the extent of the increase or decrease of the parameter under consideration. changes at the cell level and changes in the stroma tissue structure. The nuclear structure of the cell and its physical parameters such as density, elasticity, and viscosity, are changing [33], and the condensation of the cell's nucleus during apoptosis increases the ultrasound scattering [34]. On the other hand, the stromal microenvironment undergoes changes in the blood vessel architecture and the extracellular matrix composition [35,36]. In the stroma, fibrosis, collagenization, and microcalcification are associated with structures strongly scattering ultrasounds. It can be assumed that before therapy, ultrasound scattering on tumor cell clusters plays a significant role in the scattered signal. In the case of effective chemotherapy, the newly formed structures of the repair processes, such as excess fibrous connective tissues, significantly contribute to the scattering of the ultrasound; based on their size and mechanical properties, they are scatterrers influencing the value of the ENS parameter.
In one case, in the group of tumors that did not respond, a reduced ENS value after NAC treatment was found. It was an invasive NST breast cancer with in situ ductal carcinoma

Classification and statistical analysis results
Statistical analysis was performed to distinguish between response cases and no response based on ENS, IBSC, IBSC + ENS, or tumor size, features used as biomarkers for NAC response. The statistical significance of the differences between the NAC-responsive groups and the groups that did not respond was evaluated with the p-values shown in Table 1, where statistically significant differences (p < 0.05) are marked in gray. The progressive separation of responding and non-responding groups (Fig. 5) raises the question about the treatment step which allows an effective assessment of tumor response.
The QUS parameters, individually as well as their linear combination based on the LDA analysis, were used as classifiers of tumor response to NAC. The classification based on the largest tumor size was also assessed. The obtained AUC values calculated at each NAC stage are shown in Fig 6 and Table 2.   6a). A combination of QUS parameters allowed improving the AUC after the second, the third, and the fourth NAC course (Fig. 6b).
The main difference in the efficiencies of classification based on the IBSC + ENS composite classifier and the classification based on the size of tumors is the different stage of NAC in obtaining a similar AUC (Fig 6b). This difference, however, is extremely important because the early diagnosis of N-PR tumors is fundamental to personalize the treatment and allows oncologists to make an early decision depending on the effectiveness of the NAC.
For an assessment based on tumor size, often used in clinical evaluation, effective classification is only possible after NAC stage 4 and 5 (AUC = 0.75 and 0.90 respectively).
For a combined QUS classifier a comparable AUC is available after the second and third NAC courses respectively. This delay in tumor size reduction results from the lower rate of the tissue repair process, manifested by the disappearance of the tumor, compared to the changes taking place in the cells and tissue structure.

Conclusions
The results obtained suggest that ENS provides useful information to monitor NAC. Supporting information S1