**3. Results**

*3.1. Visual Assessment of the CP-OCT and C-OCE Images for Distinguishing between Non-Tumorous and Tumorous Breast Tissue*

The results based on the CP-OCT and C-OCE images for representative cases of the non-tumorous and tumorous breast tissue and differentiation among highly-aggressive breast-cancer subtypes are shown in Figures 1 and 2.

**Figure 2.** Representative depth-wise C-OCE images (**a1**–**a6**) of non-tumorous and tumorous breast tissue with corresponding histological images (**b1**–**b6**). (**a1**–**b1**) Adipose tissue with streaks of connective tissue; (**a2**–**b2**) fibroadenomatosis/fibroadenoma; (**a3**–**b3**) DCIS; (**a4**,**a5**–**b4**,**b5**) IDC of scirrhous structure (low-aggressive breast cancer subtypes); (**a6**–**b6**) IDC of solid structure (highly-aggressive breast cancer subtype). Abbreviations: A—adipose, CT—connective tissue, ADH—atypical ductal hyperplasia, FA—fibroadenomatosis, DCIS—ductal carcinoma in situ, TS—tumor stroma, TC—cluster of tumor cells.

Figure 1 shows five types of representative CP-OCT and histological images: "adipose tissue with streaks of connective tissue" (a1–c1)/"fibroadenomatosis/fibroadenoma" (a2–c2)/"DCIS" (a3–c3)/invasive low-aggressive breast cancer of scirrhous structure (a4–c4)/invasive highly-aggressive breast cancer of solid structure (a5–c5).

Benign breast tissue states are characterized by high signal-penetration depth and uniformity of the signal attenuation along the inferior border in co- and cross-polarized structural OCT images (Figure 1). The hallmark of normal adipose (fatty) tissue is a "honeycomb" structure with low sparse scattering, while fibrous structures are characterized by high uniform scattering in co- and cross-polarized structural OCT images (Figure 1(a1–c1)). Fibroadenoma is characterized by a predominance of high-intensity OCT signal in co- and cross-polarization channels (Figure 1(a2–c2)) in comparison with normal breast tissue that has a dense structure due to the presence of large fibrous collagen fibers (Figure 1(c2)).

Cases suspicious for malignancy are characterized by general reduction in signal intensity and its penetration depth, irregular inferior border. All these features cause heterogeneity of the image. In particular, DCIS (Figure 1(c3)) is characterized by the presence of localized structures with low signal intensity and clear boundaries in the surrounding fibrous stroma with a high signal intensity in the cross-polarization channel (Figure 1(b3)). In co-polarization channels DCIS is not detectable (Figure 1(a3)).

In case of invasive breast cancer, the OCT signal in the cross-polarization channel for highly-aggressive (Figure 1(b5)) and less-aggressive (Figure 1(b4)) cancer subtypes is greatly different. IDC of solid structure (highly-aggressive) demonstrates a uniform low-level OCT signal, which is associated with an increased density of tumor cells and an almost total absence of anisotropic (fibrous) structures in this tumor subtype (Figure 1(b5)). For IDC of scirrhous structure (less-aggressive subtype), the heterogeneity of the OCT signal was observed: an alternating signal of medium and low intensity was revealed (Figure 1(b4)). On the corresponding histological images, there were clusters of tumor cells surrounded by connective tissue in a state of fibrosis and hyalinosis (Figure 1(c4)), which clearly leads to an increase in the level of OCT signal in these areas. It should be noted that in these cases, there is no pronounced contrast between low-aggressive (Figure 1(a4)) and highly-aggressive (Figure 1(a5)) breast cancer subtypes in the co-polarization channel.

Thus, in the structural OCT images, the most informative is the cross- polarization channel showing both regions with fairly high cross-polarization backscattering and (corresponding to the presence of connective tissue) and regions with a reduced cross-polarization signal (corresponding to the clusters of tumor cells), see Figure 1(b1–b5)). Therefore, in view of low informativity of the co-polarization images, only cross- polarization images were used for diagnostic accuracy analysis in this study.

The C-OCE image of the normal mammary gland (normal connective tissue and adipose tissue) is characterized by the lowest stiffness (Figure 2(a1)). However, fibroadenomatosis/fibroadenoma is characterized by a slight overall increase in stiffness (Figure 2(a2)) and the presence of well-localized areas with an increased elastic modulus in the regions of atypical ductal hyperplasia (ADH).

C-OCE images of malignancy demonstrate the appearance of regions with pronouncedly increased stiffness. Moreover, for IDC of solid structure (highly-aggressive), these areas occupy up to 90% of the entire image, which sharply distinguishes this breast cancer subtype (Figure 2(a6)). The ducts filled with tumor cells for DCIS are visualized as high-contrast zones with strongly increased stiffness (Figure 2(a3)) which coincide well with the histological image. The surrounding fibrous tissue is characterized by fairly low stiffness values (Figure 2(a3)). The OCE images of IDC of scirrhous structure demonstrate an increased stiffness in the regions of the clusters of tumor cells and significantly lower stiffness in the regions of the tumor stroma, causing multiple moderately contrast inclusions with elevated stiffness, which represents a feature of low-aggressive tumor subtype (Figure 2(a4–a5)).

In addition, it is necessary to mention that images of IDC of scirrhous structure and fibroadenoma may have similar patterns that may be challenging to differentiate for the reader. To solve this problem an additional criteria (Table 3) of "presence the numerous and less contrasting inclusions of increased stiffness" was included in cases of IDC (Figure 2(a4)) in contrast to single inclusions in cases of fibroadenoma (Figure 2(a2)) and DCIS (Figure 2(a3)).

#### *3.2. Diagnostic Accuracy of CP-OCT and C-OCE Based on Visual Assessment of Images*

The results of the two tests, using the identified main and additional criteria, separately in CP-OCT images and C-OCE images demonstrate their grea<sup>t</sup> agreemen<sup>t</sup> among the readers. The concordance coe fficient in the determination of tumorous or non-tumorous breast tissue in the analysis of CP-OCT images between two researchers was k = 0.68, between two post-graduate students k = 0.93, between the two surgeons k = 0.80. The concordance coe fficient in the detection of tissue type in the analysis of C-OCE images between two researchers was k = 0.86, between two post-graduate students k = 0.93, between the two surgeons k = 0.82.

To demonstrate the variability of the test results, ROC-curves were presented for each reader (Figures 3 and 4). ROC-curves analysis confirmed that visual assessment of CP-OCT and C-OCE images has a high diagnostic value for di fferentiating non-tumorous and tumorous breast tissue (AUC values for all readers were 0.90–0.97 and 0.93–0.99, respectively) and also for distinguishing between low- and highly-aggressive invasive breast-cancer subtypes (AUC values for all readers were 0.84–0.90 and 0.80–1.00, respectively) (Figure 3c, Figure 4c). Slightly lower values were obtained for di fferentiation between non-invasive breast lession and invasive breast cancer (AUC values for all readers were 0.74–0.93 and 0.86–0.95, respectively) (Figure 3b, Figure 4b). The ROC-curves show that the best results were demonstrated by the researches experienced in optical imaging.

**Figure 3.** Receiver operating characteristic (ROC)-curves showing the results of visual assessment CP-OCT images for distinguishing non-tumorous breast tissue from tumor (**a**), DCIS from invasive breast cancer (**b**), low-aggressive invasive breast cancer from highly aggressive (**c**), non-tumorous breast tissue from low-aggressive breast cancer (**d**), non-tumorous breast tissue from highly aggressive breast cancer (**e**) for six "blinded" readers.

**Figure 4.** ROC-curves showing the results of visual assessment of C-OCE images for distinguishing non-tumorous from tumorous breast tissue (**a**), DCIS from invasive breast cancer (**b**), low-aggressive invasive breast cancer from highly-aggressive (**c**), non-tumorous breast tissue from low-aggressive breast cancer (**d**), non-tumorous breast tissue from highly-aggressive breast cancer (**e**) for six "blinded" readers.

The results of the blinded reader analysis are summarized in Table 4, showing the sensitivity, specificity and diagnostic accuracy. Each diagnostic index was averaged among all six readers. High diagnostic values were obtained for the differential diagnosis of all analyzed groups. The diagnostic accuracy of distinguishing non-tumorous tissue from tumor was 92.4 ± 2.3% for CP-OCT and 96.0 ± 3.3% for OCE, which determines the OCE method as more specific for detecting tumorous tissue.

For the first time, the diagnostic efficiency of CP-OCT and C-OCE methods for the differential diagnosis of non-invasive from invasive breast cancer was established (Se = 90.1 ± 5.7%, Sp = 70.6 ± 11.3%, Ac = 82.5 ± 7.1% and Se = 90.5 ± 5.3%, Sp = 92.0 ± 6.1%, Ac = 90.4 ± 2.7%, respectively). Furthermore, we demonstrated the possibility to differentiate invasive low-aggressive breast cancer subtypez with a favorable prognosis from highly-aggressive breast cancer subtypes with a poor prognosis for treatment and the course of the disease (Se-83.5 ± 10.5%, Sp-93.5 ± 6.0%, Ac-87.8 ± 6.5% and Se-87.3 ± 13.8 ± 6.5%, Sp-98.0 ± 3.1%, Ac-89.5 ± 10.0%, respectively). In both cases, it was demonstrated that C-OCE showed the best diagnostic indicators (Table 4).

Additionally, we performed a diagnostic analysis of the possibility to distinguish non-tumorous breast tissue from low- and highly-aggressive breast cancer subtypes. It has been shown that the diagnostic accuracy of the difference between non-tumorous breast tissue and a low-aggressive subtype of cancer is 88.1 ± 6.0% for CP-OCT and 95.7 ± 4.1% for C-OCE. The diagnostic accuracy of the difference between non-tumorous breast tissue and highly-aggressive cancer is 97.2 ± 2.8% for CP-OCT and for C-OCE—98.3 ± 2.2%.

Thus, we demonstrated the possibility to use CP-OCT and C-OCE methods for detecting different breast cancer subtypes on the resection margin which would minimize the risk of recurrence and reoperations.


**Table 4.** The results of diagnostic test for visual assessment of the CP-OCT and C-OCE images.

#### *3.3. Assessment of Human Breast Cancer Margins*

The tests performed in this study demonstrated that, in distinguishing the norm from low-aggressive cancers (and, moreover, highly aggressive ones), the analysis of both CP-OCT and C-OCE images the both methods enable high diagnostic accuracy. However, when searching for the transition between IDC of scirrhous structure and non-cancerous tissue, the C-OCE-based stiffness mapping (Figure 5c) visualizes the tumor margin much more clearly in comparison with the cross-polarization images (Figure 5b).

**Figure 5.** Histological image (**a**) demonstrating transition between non-tumorous (fibrous stroma—FS) and tumorous breast tissues (low-aggressive IDC of scirrhous structure); (**b**) is the corresponding CP-OCT image in the cross-polarization channel and (**c**) is the C-OCE images of the same area. HS denotes hyalinized stroma, and TC—clusters of tumor cells.

## **4. Discussion**

The results presented here show the high diagnostic value and e fficiency of CP-OCT and C-OCE methods for di fferential diagnosis of non-tumorous and tumorous breast tissue, with the further prospect of intraoperative determination of the "positive" margin of tumor resection during breast-conserving surgery in real time. In addition, the diagnostic e fficiency of CP-OCT and C-OCE methods for di fferentiation between non-invasive and invasive breast cancers, as well as between invasive low-aggressive breast cancer subtype with a favorable prognosis (Luminal A, Luminal B (Her2/Neo-)) and highly aggressive breast cancer subtypes with a poor prognosis for course of the disease (Her2/Neo+, Non-luminal, TNC).

In previous studies, only standard visual imaging criteria, such as signal intensity and high/low sti ffness, were used for di fferentiation between tumorous and non-tumorous breast tissues. In this study, additional analysis criteria were proposed, which made it possible to increase the diagnostic sensitivity and specificity, significantly reducing the number of erroneous diagnoses. We identified such additional analysis criteria as the presence of structures and the characteristics of signal attenuation in depth on cross-polarization images, as well as the presence of inclusions and mosaic structure on C-OCE images with visually feasible assessment of several sti ffness grades.

Previous works have demonstrated that conventional, intensity-based OCT can provide di fferentiation between tumorous and non-tumorous breast tissues through both quantitative [18,61–63] and qualitative [32,61,63] assessment of the OCT signal. Several studies demonstrated that OCE has the high potential to delineate tumor in breast tissue based on elevated elasticity on a microscale [33,40,41,44]. A recent study [16] demonstrated the ability of structural OCT to identify positive margins in specimens from BCS. The qualitative assessment of OCT images showed the high diagnostic accuracy of structural OCT for distinguishing normal and cancerous tissue within the resection bed following wide local excision of the human breast: sensitivity of 91.7% and specificity of 92.1% [16]. Additionally, visual assessment of C-OCE images for evaluation of tumor margins in specimens excised during breast-conserving surgery also provides high sensitivity of (92.9%) and specificity (96.4%) [43].

Breast cancer is a highly heterogeneous disease, both morphologically and genetically. The surgical approach and the amount of resection depend on the subtype of breast cancer, which, as this study has shown, can be determined in rapid OCT-based tests, including the possibility of intraoperative use. The C-OCE and CP-OCT images provide additional contrast between tumor and normal tissue in comparison with structural OCT. C-OCE and CP-OCT analysis of excised tissue specimens can distinguish between normal and cancerous tissues by identifying the heterogeneous and disorganized connective tissue structures indicative for malignancy. We have demonstrated that di fferences in the microstructural features of cross-polarization and sti ffness images enable di fferentiation between highly and low-aggressive breast cancer subtypes confirmed by histopathology. For this purpose, the main and additional criteria for assigning an image to a particular group were formulated, which are necessary for a more accurate di fferentiation of malignant conditions among themselves. For example, a uniform low-intensity in CP-OCT images and a uniform high level of sti ffness in C-OCE images characterize tumor of a solid structure, while tumor tissue of a scirrhous structure in the immediate vicinity of non-tumorous breast tissue can also retain homogeneity, or it can lose it and may be represented by di fferent levels of signal intensity and sti ffness.

ROC curves were constructed as a measure of overall accuracy for each reader when non-cancerous tissue was distinguished from tumor by CP-OCT and C-OCE methods (Figure 3). In this case, C-OCE showed higher specificity (97.5 ± 2.7% vs. 93.3 ± 6.0%) and diagnostic accuracy (96.0 ± 3.3% vs. 92.4 ± 2.3%) compared to cross-polarized images. This fact may be caused by the di fficulty in interpreting qualitative OCT criteria based on signal intensity by readers, in comparison with the criteria for interpreting quantitative OCE images that usually have more contrast and visually easier assessable di fferences. Overall, for di fferentiation between tumorous and non-tumorous tissues, the C-OCE method has proved to be more e fficient.

Additionally, for more specific differentiation between non-invasive breast cancer and invasive breast cancer, the following diagnostic parameters were determined for CP-OCT and C-OCE methods: Se = 90.1 ± 5.7%, Sp = 70.6 ± 11.3%, Ac = 82.5 ± 7.1% and Se = 90.5 ± 5.3%, Sp = 92.0 ± 6.1%, Ac = 90.4 ± 2.7%, respectively. For distinguishing between invasive low-aggressive and highly-aggressive breast cancer subtypes, the CP-OCT and C-OCE gave the following results: Se = 83.5 ± 10.5%, Sp = 93.5 ± 6.0%, Ac = 87.8 ± 6.5% and Se = 87.3 ± 13.8 ± 6.5%, Sp = 98.0 ± 3.1%, Ac = 89.5 ± 10.0%, respectively. Therefore, in both cases, C-OCE showed better diagnostic indicators (Table 4).

The diagnostic accuracy of the di fference between non-tumorous breast tissue and low-aggressive breast cancer for CP-OCT and C-OCE was found to be fairly high, 88.1 ± 6.0% and 95.7 ± 4.1%, respectively. Even higher was the Ac of CP-OCT and C-OCE for the di fference between non-tumorous breast tissue and highly-aggressive breast cancer (97.2 ± 2.8% and 98.3 ± 2.2%, respectively).

Accurate determining of the boundaries of tumor resection is more feasible for tumors of a solid structure in comparison with tumors of scirrhous structure that may resemble fibroadenomas in OCT-based images. However, the performed targeted histological examination has given a clue for better understanding of the causes of sti ffness increase or decrease and made it possible to define additional criteria that improved the diagnostic accuracy of C-OCE for various breast cancer subtypes detection, including non-invasive and low-aggressive tumors.

Thus, the formulated additional (clarifying) criteria for visual assessment of CP-OCT and C-OCE images provided a higher diagnostic accuracy in di fferentiation between tumorous and non-tumorous breast tissues with various grades of aggressiveness. In the future, this will increase the value of these OCT-based methods in detecting the boundaries of tumor resection during BCS.
