Open Access
Add to BibSonomy
Abstract
Early detection of breast cancer can significantly improve patient outcomes and five-year survival in clinical screening. Dynamic optical breast imaging (DOBI) technology reflects the blood oxygen metabolism level of tumors based on the theory of tumor neovascularization, which offers a technical possibility for early detection of breast cancer. In this paper, we propose an intelligent scoring system integrating DOBI features assessment and a malignancy score grading reporting system for early detection of breast cancer. Specifically, we build six intelligent feature definition models to depict characteristics of regions of interest (ROIs) from location, space, time and context separately. Similar to the breast imaging-reporting and data system (BI-RADS), we conclude the malignancy score grading reporting system to score and evaluate ROIs as follows: Malignant (≥ 80 score), Likely Malignant (60−80 score), Intermediate (35−60 score), Likely Benign (10-35 score), and Benign (<10 score). This system eliminates the influence of subjective physician judgments on the assessment of the malignant probability of ROIs. Extensive experiments on 352 Chinese patients demonstrate the effectiveness of the proposed system compared to state-of-the-art methods.
© 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Breast cancer is one of the most common malignant tumors in women [1]. In recent years, there has been an increase in the number of breast cancer patients in younger age groups [2]. Studies show that breast cancer treatment especially in the early stage of diseases is exceptionally effective, with relative reductions in mortality between 15% and 25% in randomized trials [3–5]. Therefore, superior methods for early screening and detection of breast cancer are essential to improve the prognosis and outcome of the disease.
Nowadays, many imaging modalities have been employed to assist in breast cancer screening and detection, mainly falling into two categories: morphological imaging and functional imaging [6–9]. As the predominant morphological imaging modality, mammography (MG) has been shown to reduce breast cancer mortality by 15% to 45% but has a diminished sensitivity on radio-graphically dense breasts [10–12]. Ultrasound (US), is widely used for diagnostic imaging and is advocated for supplementary screening in average-risk women with dense breasts [13–15]. Functional imaging modalities have been demonstrated to have higher incremental cancer detection rates in supplementary screening of dense breasts, such as traditional magnetic resonance imaging (MRI) [16–19]. However, these imaging modalities are disadvantaged by over-diagnosis or low positive predictive values [20–22]. MG is sensitive to detecting large lesions and fail to capture breast cancer lesions at an early stage. Although US is well known to detect smaller invasive cancer, it has a low detection rate for carcinoma in situ and is not easy to show small lesions or cannot reliably distinguish benign and malignant lesions. MRI usually is detrimental to patients, causing over-diagnosis in early large-scale screening [23–25].
Benefiting from the development of imaging technology, dynamic optical breast imaging (DOBI) technology has been expected to capture features early in tumorigenesis, which can theoretically detect tumors as small as 2 mm based on tumor neovascularization theory [26–34]. Several studies have shown that DOBI is effective in young women and shows promise for discriminating malignancies from benign lesions in a low-cost, non-invasive manner [34–38]. However, current DOBI-based studies have a preliminary analysis of the breast characteristics in European and American women but do not perform well on Chinese women’s breasts. One of the main reasons may be that Chinese women’s breasts are smaller and denser than those of European and American women [39–43], resulting in significant differences in breast lesion imaging or blood oxygen concentration levels. Besides, inadequate lesion features or inappropriate selection of background areas also lead to poor detection performance.
Aiming at above problems, we proposed and validated DOBI-based Intelligent Scoring System (DOBI-ISS) for early detection of breast cancer. The two core components are features definition, including the location, space, time, and context of ROIs and a malignancy score grading reporting system like BI-RADS. We applied DOBI-ISS to analysis of 352 Chinese women and compared it with state-of-the-art DeHCA [36], ultrasound, and mammography. Extensive experiments showed that DOBI-ISS could be used as a complement to ultrasound or mammography for early detection of breast cancer.
2. Materials and methods
2.1 Datasets
In this paper, we collected DOBI data on 352 patients between August 2017 and October 2022 from four hospital institutions in different regions of China, which were as follows: (i) Sir Run Run Shaw Hospital Zhejiang University School of Medicine, Zhejiang Province; (ii) The First Affiliated Hospital of Xi’an Jiaotong University, Shannxi Province; (iii) Sun Yat-sen Memorial Hospital, Sun Yat-sen University, Guangdong Province; and (iv) Zhejiang Cancer Hospital, Zhejiang Province. All experiments were approved by the relevant Ethics Committee. The written agreement was given by all patients before collecting data. All data were imaged by the DOBI Medical ComfortScan System and met the inclusion criteria [34–38]. DOBI Medical ComfortScan System is an optical mammography imaging system based on DOBI and designed to detect dynamic changes of increased blood volume levels and depleted oxygen levels, that characterize malignancies. It is made of three physical components: the optical system, the soft breast locator, and the charge-coupled device (CCD) acquisition system with a resolution (768×512 pixels). In this study, there are two differences from the device used in previous studies [34–38]: (i) The optical system improves the LED panel by changing the number and locations of LEDs, which makes it more suitable for the physical characteristics of Chinese patients. (ii) The LED panel emits light at a wavelength of 640 nm, complemented by 800 nm light for imaging enhancement, which helps reduce imaging noise. Labels for location, space, time, and context (LSTC) features of ROIs were produced by two trained and experienced radiologists. Of the pathologically confirmed cases, 208 were benign and 144 were malignant. The time between DOBI data acquisition and pathology diagnosis in all patients was no more than 4 weeks. In addition, some cases had ultrasound or mammography results diagnosed by two radiologists with 20 years of experience during the same period, with 174 patients having ultrasound results and 77 patients having mammography results. The results of US and MG are analyzed using the BI-RADS criteria. In particular, each case had only one ROI or suspicious area (SA) in this study.
2.2 Methods
The proposed DOBI-ISS consists of four steps, as shown in Fig. 1: (i) Regions of interest (ROIs) extraction, (ii) Reference ROIs extraction based on a multi-background strategy, (iii) LSTC features definition by machine learning algorithms, and (iv) DOBI-Scores calculation and DOBI-Levels assignment for ROIs.
2.2.1 Regions of interest (ROIs) extraction
ROIs extraction from breast tissues is key to differentiating malignant from benign lesions. ROIs are the areas where tumors that grow with angiogenesis support have higher metabolic rates and produce more deoxyhemoglobin than normal tissues. Since deoxyhemoglobin is sensitive to light at 640 nm, it generally appears as a dark blue light-absorbing region in the image [34]. Therefore, we consider dark blue areas in a series of sequential DOBI data as ROIs or SAs and design an algorithm to extract these ROIs. Detailed ROIs extraction processes are provided in the Supplement 1.
2.2.2 Reference ROIs extraction based on a multi-background strategy
An appropriate reference ROI that is not affected by the conditions being assessed in the breast, is essential because comparing differences between the ROI and the reference assists with assessment the malignant probability of the ROI. Unlike previous studies with one reference ROI, we present a multi-background strategy in Fig. 2 that selects multiple regions with consistent background behaviors to determine the reference ROI. Specifically, we first remove the ROIs and areas influenced by conditions being assessed in the breast imaging region. The rest are clustered into K subparts by Bisecting K-means, and then K temporal curves reflecting changes in light absorption content come out. If more than half of these K temporal curves have similar patterns, indicating that the background is consistent “behavior”, then the average of these similar temporal curves is regarded as a temporal curve of true reference ROI. Otherwise, the background is “behaving” inconsistently, there is no appropriate reference ROI, and it needs to be dealt with separately.
Fig. 2. The processes of reference regions of interest (ROIs) extraction. The dynamic optical breast image at the moment of deepest ROI absorption (a), the remaining part of image after removing the ROI (b), ten subparts (representing ten reference ROIs) clustered by Bisecting Kmeans (c), and ten temporal curves reflecting changes in light absorption content or metabolic rate (d).
2.2.3 LSTC features definition by machine learning algorithms
We characterize ROIs from four features: location, space, time and context, which embody both morphological and metabolic characteristics of breast cancer lesions over time.
Location
The location is used to assess ROI proximity to the acquisition marker (AM) corresponding to the area of concern identified by prior clinical or mammography findings. If the ROI is closer to the AM, it is more likely to be malignant. Here we consider breast sizes and measure the distance from the ROI to the AM to weigh the malignant probability of the ROI.
Space
From the appearance of ROIs, we define the space from three aspects as shown in Fig. 3: peak or bowl, focus or diffuse, and stable or wave. The peak-like or bowl-like feature mainly considers the depth of the absorption transition from the ROI to its surrounding regions. If ROI margins have relatively steep transitions (peak-like, Fig. 3(a)), it is more likely to be associated with malignancy. In contrast, if the ROI looks bowl-like (some transition but with a large, even-colored area of deep blue, Fig. 3(b)), it is not clearly benign. The focus or diffuse feature also reflects “transition from adjacent areas” and “intensity of ROIs”. An intense focal ROI is likely considered a malignancy (Fig. 3(c)), whereas a very diffuse blue area is not consistently associated with malignant or benign findings (Fig. 3(d)). The spatial stability reflects epicenter positions of the ROI during dynamic sequences. If the epicenter of the ROI stays in one location during sequences (“stable”, Fig. 3(e)), it is more often associated with malignancy.
Fig. 3. Cases of different spatial characteristics of regions of interest (ROIs) in space: peak-like (a), bowl-like (b), focus (c), diffuse (d), stable (e), and wave (f). The red point represents the acquisition marker, the yellow point represents the epicenter of the ROI, and the blue point represents the nipple in each image at the moment of deepest ROI absorption.
In a word, a peak-like, focal, intense, and stable ROI is likely to be malignant. Conversely, very diffuse or wavy blue areas are not consistently associated with either malignant or benign findings. We employ three machine learning algorithms to define these features: an Adaboost classifier to classify ROI shapes into peaks or bowls, an L2 penalized logistic regression model to determine whether the ROI is focused or diffuse, and an SVM model with an RBF kernel to classify the stability of the ROI into stable or wave. More details are available in the Supplement 1.
Time
The characteristic of time reflects changes in blood oxygen metabolism over time in ROI, depicted as the temporal curve. We divide temporal curves into four categories by shape to reflect the malignant probability of the ROI, as shown in Fig. 4. The S-1 curve, positive over time and upward, is consistent with the benign region. The S0 curve, wavy and sinusoidal or highly variable in amplitude, is not clearly associated with a benign or malignant region. The S1 curve, a moderately wavy line but with a strong downward trend and a net change, is considered likely malignant. The S2 curve is as strongly malignant as a straight line with the same downward trend. Detailed definition model is present in the Supplement 1.
Context
The context is used to evaluate the similarity between time feature of the ROI and the reference ROI. The smaller the distance between the two curves and the more consistent the line shapes, the more similar. At this point, the ROI is likely benign. We characterize the ROI context as similar, dissimilar, and indeterminate, which are described in the Supplement 1.
2.2.4 DOBI-Score & DOBI-Level:
To quantify the malignant probability of the ROI, we score and summarize LSTC features to obtain DOBI-Score that characterizes how malignant the ROI is. The DOBI-Score is defined as follows:
A high score of the ROI indicates a high malignant probability, while a low score is often associated with benign conditions. Moreover, with reference to BI-RADS, we give a grade and meaning of malignancy in ROIs from benign, likely benign, intermediate, likely malignant, and malignant in Table 1, called DOBI-Level.
Table 1. The definition of DOBI-Level of regions of interest (ROIs)
3. Results
3.1 Environment and parameters
Experiments were performed on an NVIDIA GeForce RTX 3090 GPU with a Python 3.9 environment. Parameter settings of feature definition models were available in the Supplement 1. When calculating DOBI-Scores for ROIs, we first assigned different weights to each feature dimension in Table 2 by combining the physician empirical weights and feature importance, which were described in the Supplement 1. The higher the DOBI scores, the higher the likelihood of malignancy for ROIs, and the more malignant the corresponding DOBI-Level. Then we divided the DOBI-Score into different levels: [0, 10] as benign, (10, 35] as likely benign, (35, 60] as Intermediate-Indeterminate, (60, 80) as likely malignant, and [80, 100] as malignant by data analysis. We also gave an example analyzed by our system in Fig. 5.
Fig. 5. Dynamic optical breast image (DOBI) visualization of a case at the moment of deepest ROI absorption (a) and a metabolic curve of light absorption (b). Using DOBI-ISS, Location: Yes, 7 scores; Space: stable, foucs, peaked, and 23 scores; Time: S2 curve, 38 scores; Context: dissimilar, 23 scores. The DOBI-score of the ROI was calculated as 91, and the DOBI-Level was regarded as malignant (M). In particular, the biopsy result was invasive ductal carcinoma.
Table 2. The table of feature scores
3.2 Models analysis
To verify the effectiveness of feature definition models in proposed DOBI-ISS, we record and calculate accuracy, precision, sensitivity, Area Under the Curve (AUC), and F1-score, as shown in Table 3. It is important to note that these AUC values are only the performance of the machine learning algorithms involved in these feature extractors. As can be seen from Table 3, our models were able to accurately identify categories with different features on raw DOBI data. We also add a comparison of the performance of our feature extractors and the general feature extraction method based on neural networks on experimental data in the Supplement 1. The features identified by these six models were 92% consistent with physicians. For misjudged samples, we found that the background areas of most samples were disordered. It should also be noted that Space_shape may require a transitional classification description rather than an absolute judgment like a peak or bowl. In a word, our models are able to label features reasonably based on their importance, indicating that DOBI-ISS can be better model physicians and thus achieve intelligent interpretation of DOBI data.
Table 3. The performance of models for six feature definitions
In order to analyze the plausibility of DOBI-Score and DOBI-Level, we first calculate the positive predictive value (PPV) and negative predictive value (NPV) corresponding to different DOBI-Levels. The results are shown in Table 4. From Table 4, it can be seen that as the DOBI-Score increases and the malignant probability increases, the PPV becomes greater and the NPV becomes less, which makes sense to illustrate the DOBI-Level definition. Besides, we also performed some experiments to convert BI-RADS results from ultrasound and mammogram into the DOBI-Level in the Supplement 1, verifying the effectiveness of DOBI-level definition.
Table 4. The results of the proposed system a
3.3 Comparison with the existing methods
We compared the performance of DOBI-ISS with that of state-of-the-art method DeHCA [36] on DOBI data from 352 patients. Since contralateral breast information was required to calculate DeHCA scores, 49 of 352 patients with dual breast DOBI information were applied to the DeHCA method and DOBI-ISS, respectively. The DOBI-ISS had a high sensitivity of 0.9 and a high accuracy of 0.69 when score threshold was 58, while DeHCA method had the sensitivity of 0.14 and accuracy of 0.57, with the referred score threshold 0.85. As a result, DeHCA prefers to assess the tumor as benign, leading to missed calls, however our system can detect more malignancies with a higher F1-score. Meanwhile, our DOBI-ISS has a more detailed malignant grades than DeHCA and is more suitable for breast cancer screening. More detailed comparison results are depicted in the Supplement 1.
We also compared the diagnostic performance of US, MG, and DOBI-ISS, filtered cases with absent US or MG results, and collected 69 cases for fairly comparison analysis. Here we considered BI-RADS 4b, BI-RADS 4c, BI-RADS 5, BI-RADS 6, and DOBI-ISS with DOBI scores >58 (some DOBI-Level I, DOBI-Level LM, and DOBI-Level M) as malignant, and the rest as benign. The comparison results were depicted in Table 5. DOBI-ISS had the best sensitivity of 0.98, which far exceeded the results of US and MG. DOBI-ISS called the most malignant cases but it was mixed with more benign cases than others, resulting in a slightly inferior specificity. US had the best specificity of 0.94, which indicated that the probability of misdiagnosis being malignant was the lowest. From the overall performance analysis, DOBI-ISS had the best F1-score and accuracy, followed by US, and MG had a slightly inferior F1-score and accuracy.
Table 5. The comparison results of 69 patients by the proposed system, ultrasound, and mammography, respectively
In order to analyze the consistency and validity of comparison results, we displayed overlaps of all cases, benign, and malignant detected by these four methods (DOBI-ISS, US, MG, and biopsy) as shown in Fig. 6(a), (b), and (c), respectively. Of all benign and malignant cases, DOBI-ISS had the most overlaps with biopsy, followed by the US, and the lowest mammography. In malignant cases, we had the same conclusion, indicating that DOBI-ISS had a more robust performance especially on calling malignant tumors.
Fig. 6. The Venn charts of results of four methods: the proposed system, ultrasound (US), mammography (MG), and biopsy. (a) overlaps of all cases. (b) overlaps of benign cases. (c) overlaps of malignant cases.
DOBI-ISS has high overlaps with ultrasound (55/69) and mammography (49/69), and can also call some malignant cases which cannot be detected by ultrasound or mammography. Of 69 cases, 10 cases (benign: malignant = 1:9) were detected errors by ultrasound. But all 10 cases were detected as malignant by DOBI-ISS, which showed that DOBI-ISS could detect more malignant tumors than ultrasound. We get the same conclusion on the results of mammography. Therefore, the proposed DOBI-ISS can be used as a supplementary tool to ultrasound and mammography, especially in calling malignancies in the early stages of breast cancer.
As a result of the biopsy, there were four malignant cases that only DOBI-ISS detected, but US and MG could not detect. Of the four malignant cases, three were diagnosed as invasive ductal carcinomas and one was breast adenopathy with lobular neoplasia by biopsy. Two of these four cases with tumors less than 15 mm in the US were small cups. In US results, these four cases were described as hypoechoic, morphologically regular, clear side cuts, smooth edges, uniform echo, and inconspicuous blood flow. In MG results, three cases were punctate calcifications and one case did not show obvious sediment-like calcifications, and all cases were diagnosed as breast hyperplasia or nodular masses, benign. Using DOBI-ISS, we described ROIs of these four cases as peaked, stable, focused, S1 or S2 curves, and dissimilar to the background, which belonged to tumors with strong malignancy.
There were five cases, including four benign and one malignancy, which were not detected correctly by DOBI-ISS, but correctly detected by MG and one miscalled by US. By analyzing the LSTC results, we inferred that the capillary metabolic rate of benign tumors in the early stages was not significantly more obvious than that of malignant tumors in DOBI images. The current four features of LSTC might not be perfect for distinguishing benign and malignant tumors. Fortunately, if we combine the results of DOBI-ISS with those of US or MG, there would be a higher performance in breast cancer detection than only using one method.
In conclusion, our DOBI-ISS can detect many malignancies in the early stages of breast cancer and can be used as a supplement to ultrasound or mammography.
4. Discussion
Unlike most algorithms based on ultrasound or mammography data [44–46], this study analyzed characteristics of ROIs in breast issues and explored the feasibility of the proposed system based on DOBI data from Chinese women. In this paper. we built DOBI-ISS to automatically delineate the location, space, time, and context features of ROIs after ROIs extraction, and then calculated the DOBI-Score and graded DOBI-Level for characterization the malignant probability of ROIs. Suppose a mass is focal, intense, stable, peaked (namely, the margins have relatively steep transitions), with the time curve of the downtrend (namely, the concentration of neovascular deoxyhemoglobin gradually increases), and its time curve diverges from curves of consistent background behaviors (namely, the level of oxygen metabolism in this area is not consistent with the surrounding normal area), the mass is considered a malignant lesion with a high DOBI-score. To better automate the identification of different dimensional characteristics of ROIs, we not only considered rule constraints from empirical experts, but also built corresponding classification models by mining statistical features. Experiments showed that DOBI-ISS had a high PPV with a high level of malignancy, which was very plausible. The accuracy of DOBI-ISS was up to 0.69 and surpassed the accuracy of previous work DeHCA (0.57), showing the potential of DOBI-ISS. Additionally, the comparison results with ultrasound and mammography showed that the proposed DOBI-ISS could detect additional malignancies and be used as a supplement to ultrasound or mammograms for detection in the early stages of breast cancer.
Currently, there were also some limitations associated with our study. First, some features were poorly classified with only two or three classification results, especially the feature of space or the context. For example, it should be noted that the feature of space shapes might require a transitional classification description rather than an absolute judgment like a peak or bowl. Meanwhile, the LSTC features were inadequate, and some biopsy benign cases had malignant LSTC features, which introduced unnecessary false positive cases. Therefore, it is necessary to divide features more granularly and explore more interpretable features from different dimensions. In addition, all features were analyzed based on one ROI lesion of the breast in this paper, regardless of the interaction between multiple ROI lesions. Multiple ROIs-based studies would be proposed in the follow-up studies for more accurate and interpretable evaluation of tumors. Moreover, the selection process of hyperparameters in feature extraction models is slightly coarse. In the future, we plan to design more appropriate methods to choose these hyperparameters based on more collected data.
In summary, we show that DOBI-ISS can accurately characterize malignancies and automatically score regions of interest in breast to eliminate the influence of subjective physician judgments, which is important for early clinical screening and diagnosis of breast cancer.
5. Conclusions
For early detection of breast cancer based on DOBI technology, an intelligent scoring system, named DOBI-ISS, was proposed to assess the malignant degree of tumors. Our approach extracted and described ROIs from the location, space, time, and context features and designed a malignancy score grading reporting system to evaluate the malignant probability of ROIs. To objectively depict these features of ROIs, we mined some statistical characteristics and employed suitable machine-learning algorithms with clinical experiences, mitigating the influence of subjective physician judgments. To facilitate interpretability and clinical decision support, we provided a malignancy score grading reporting system like BIRADS.
In conclusion, this study presents an intelligent scoring system for dynamic optical breast imaging in automatic image diagnosis. It offers a new tool for early clinical screening and diagnosis of breast cancer. We have demonstrated its validity and limitations through extensive analyses. Ultimately, it must be demonstrated through more patients from different regions. Verifying its practical usefulness on more datasets and optimizing its performance also remains an area of future research.
Acknowledgments
We would like to thank Guowang. John Zhang (from DOBI Medical) for providing DOBI data support and technical guidance, Xiaojia Wang (from Zhejiang Cancer Hospital) and Xiaowei Wang (from DOBI Medical Clinic) for the radiological image interpretation, and all patients who provide their clinical information for this study.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
Supplemental document
See Supplement 1 for supporting content.
References
1. A. Vourtsis and W. A. Berg, “Breast density implications and supplemental screening,” Eur Radiol 29(4), 1762–1777 (2019). [CrossRef]
2. M. Aine, C. Boyaci, J. Hartman, et al., “Molecular analyses of triple-negative breast cancer in the young and elderly,” Breast Cancer Res. 23(1), 20 (2021). [CrossRef]
3. T. Wang, P. Li, Q. Qi, et al., “A multiplex blood-based assay targeting DNA methylation in PBMCs enables early detection of breast cancer,” Nat. Commun. 14(1), 4724 (2023). [CrossRef]
4. Y. Wei, P. Jasbi, X. Shi, et al., “Early breast cancer detection using untargeted and targeted metabolomics,” J. Proteome Res. 20(6), 3124–3133 (2021). [CrossRef]
5. F. Cardoso, S. Kyriakides, S. Ohno, et al., “Early breast cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up,” Ann Oncol 30(10), 1674 (2019). [CrossRef]
6. Q. Zhu and S. Poplack, “A review of optical breast imaging: Multi-modality systems for breast cancer diagnosis,” Eur. J. Radiol. 129, 109067 (2020). [CrossRef]
7. K. L. Britt, J. Cuzick, and K. A. Phillips, “Key steps for effective breast cancer prevention,” Nat. Rev. Cancer 20(8), 417–436 (2020). [CrossRef]
8. A. Conti, A. Duggento, I. Indovina, et al., “Radiomics in breast cancer classification and prediction,” Semin. Cancer Biol. 72, 238–250 (2021). [CrossRef]
9. R. Gerami, S. Sadeghi Joni, N. Akhondi, et al., “A literature review on the imaging methods for breast cancer,” Int J Physiol Pathophysiol Pharmacol 14(3), 171–176 (2022).
10. S. Di Maria, S. Vedantham, and P. Vaz, “X-ray dosimetry in breast cancer screening: 2D and 3D mammography,” Eur. J. Radiol. 151, 110278 (2022). [CrossRef]
11. W. Lotter, A. R. Diab, B. Haslam, et al., “Robust breast cancer detection in mammography and digital breast tomosynthesis using an annotation-efficient deep learning approach,” Nat. Med. 27(2), 244–249 (2021). [CrossRef]
12. A. Yala, C. Lehman, T. Schuster, et al., “A deep learning mammography-based model for improved breast cancer risk prediction,” Radiology 292(1), 60–66 (2019). [CrossRef]
13. J. Geisel, M. Raghu, and R. Hooley, “The role of ultrasound in breast cancer screening: the case for and against ultrasound,” Semin Ultrasound CT MR 39(1), 25–34 (2018). [CrossRef]
14. R. M. Mann, A. Athanasiou, P. A. T. Baltzer, et al., “Breast cancer screening in women with extremely dense breasts recommendations of the European Society of Breast Imaging (EUSOBI),” Eur. Radiol. 32(6), 4036–4045 (2022). [CrossRef]
15. R. Guo, G. Lu, B. Qin, et al., “Ultrasound imaging technologies for breast cancer detection and management: a review,” Ultrasound Med Biol 44(1), 37–70 (2018). [CrossRef]
16. W. A. Berg, “Breast MRI for “the masses”,” Eur. Radiol. 32(6), 4034–4035 (2022). [CrossRef]
17. D. Miladinova, “Molecular imaging in breast cancer,” Nucl Med Mol Imaging 53(5), 313–319 (2019). [CrossRef]
18. A. I. Huppe, A. K. Mehta, and R. F. Brem, “Molecular breast imaging: a comprehensive review,” Semin Ultrasound CT MR 39(1), 60–69 (2018). [CrossRef]
19. K. Coffey and M. S. Jochelson, “Contrast-enhanced mammography in breast cancer screening,” Eur. J. Radiol. 156, 110513 (2022). [CrossRef]
20. N. Aristokli, I. Polycarpou, S. C. Themistocleous, et al., “Comparison of the diagnostic performance of Magnetic Resonance Imaging (MRI), ultrasound and mammography for detection of breast cancer based on tumor type, breast density and patient's history: A review,” Radiography 28(3), 848–856 (2022). [CrossRef]
21. C. E. Comstock, C. Gatsonis, G. M. Newstead, et al., “Comparison of Abbreviated Breast MRI vs Digital Breast Tomosynthesis for Breast Cancer Detection Among Women With Dense Breasts Undergoing Screening,” Jama 323(8), 746–756 (2020). [CrossRef]
22. X. Liu, J. Zhang, J. Zhou, et al., “Multi-modality radiomics nomogram based on DCE-MRI and ultrasound images for benign and malignant breast lesion classification,” Front. Oncol. 12, 992509 (2022). [CrossRef]
23. L. Green, “What Is the Best Imaging Modality for Breast Cancer Detection in Women with a Personal History?” Acad Radiol 29(10), 1466–1468 (2022). [CrossRef]
24. E. Michaels, R. O. Worthington, and J. Rusiecki, “Breast cancer: risk assessment, screening, and primary prevention,” Med. Clin. North Am. 107(2), 271–284 (2023). [CrossRef]
25. T. Ugai, N. Sasamoto, H. Y. Lee, et al., “Is early-onset cancer an emerging global epidemic? Current evidence and future implications,” Nat. Rev. Clin. Oncol. 19(10), 656–673 (2022). [CrossRef]
26. J. Folkman, “Angiogenesis in cancer, vascular, rheumatoid and other disease,” Nat. Med. 1(1), 27–30 (1995). [CrossRef]
27. S. Li, H. X. Xu, C. T. Wu, et al., “Angiogenesis in pancreatic cancer: current research status and clinical implications,” Angiogenesis 22(1), 15–36 (2019). [CrossRef]
28. D. Ribatti and F. Pezzella, “Overview on the different patterns of tumor vascularization,” Cells 10(3), 639 (2021). [CrossRef]
29. P. Mehnati, S. Khorram, M. S. Zakerhamidi, et al., “Near-Infrared visual differentiation in normal and abnormal breast using hemoglobin concentrations,” J Lasers Med Sci 9(1), 50–57 (2017). [CrossRef]
30. S. G. Kandlikar, I. Perez-Raya, P. A. Raghupathi, et al., “Infrared imaging technology for breast cancer detection–Current status, protocols and new directions,” Int. J. Heat Mass Transfer 108, 2303–2320 (2017). [CrossRef]
31. J. Folkman, “Tumor angiogenesis: therapeutic implications,” N Engl J Med 285(21), 1182–1186 (1971). [CrossRef]
32. G. Bergers and L. E. Benjamin, “Tumorigenesis and the angiogenic switch,” Nat. Rev. Cancer 3(6), 401–410 (2003). [CrossRef]
33. G. Zhang, W. Wang, and H. Jiang, “DYNAMIC OPTICAL BREAST IMAGING - A New Technique to Detect Breast Cancer at Early Stage - Results of Investigational Use of Dobi Comfortscan In China,” http://dobiglobal.com/pdfs/DOBI-ANewTechniqueToDetectBreastCancerInEarlyStage_ENG.pdf.
34. A. Athanasiou, D. Vanel, C. Balleyguier, et al., “Dynamic optical breast imaging: a new technique to visualise breast vessels: comparison with breast MRI and preliminary results,” Eur. J. Radiol. 54(1), 72–79 (2005). [CrossRef]
35. M. D’Aiuto, G. Frasci, M. L. Barretta, et al., “The dynamic optical breast imaging in the preoperative workflow of women with suspicious or malignant breast lesions: development of a new comprehensive score,” ISRN Oncol 2012, 1–9 (2012). [CrossRef]
36. A. Petrillo, O. Catalano, R. Fusco, et al., “Optical imaging of the breast: evaluation of deoxyhemoglobin concentration alteration in 166 patients with suspicious breast lesions,” Eur Radiol Exp 2(1), 8 (2018). [CrossRef]
37. W. Yang, Z. Wang, K. Chen, et al., “A comprehensive classification system for breast cancer diagnosis based on dynamic optical breast imaging,” Annu Int Conf IEEE Eng Med Biol Soc 2019, 4741–4744 (2019). [CrossRef]
38. L. S. Fournier, D. Vanel, A. Athanasiou, et al., “Dynamic optical breast imaging: a novel technique to detect and characterize tumor vessels,” Eur. J. Radiol. 69(1), 43–49 (2009). [CrossRef]
39. J. M. Bae and E. H. Kim, “Breast density and risk of breast cancer in Asian women: a meta-analysis of observational studies,” J Prev Med Public Health 49(6), 367–375 (2016). [CrossRef]
40. W. Ren, M. Chen, Y. Qiao, et al., “Global guidelines for breast cancer screening: A systematic review,” Breast 64, 85–99 (2022). [CrossRef]
41. J. Liu, P. F. Liu, J. N. Li, et al., “Analysis of mammographic breast density in a group of screening Chinese women and breast cancer patients,” Asian Pac J Cancer Prev 15(15), 6411–6414 (2014). [CrossRef]
42. D. W. Lim, V. Giannakeas, and S. A. Narod, “Survival differences in Chinese versus white women with breast cancer in the United States: a SEER-based analysis,” JCO Glob Oncol 6(6), 1582–1592 (2020). [CrossRef]
43. Y. Wang, Y. Li, Y. Song, et al., “Comparison of ultrasound and mammography for early diagnosis of breast cancer among Chinese women with suspected breast lesions: A prospective trial,” Thorac. Cancer 13(22), 3145–3151 (2022). [CrossRef]
44. A. B. Nassif, M. A. Talib, Q. Nasir, et al., “Breast cancer detection using artificial intelligence techniques: A systematic literature review,” Artif Intell Med 127, 102276 (2022). [CrossRef]
45. R. Fusco, V. Granata, and A. Petrillo, “Introduction to Special Issue of Radiology and Imaging of Cancer,” Cancers 12(9), 2665 (2020). [CrossRef]
46. J. Zhang, L. Cai, X. Pan, et al., “Comparison and risk factors analysis of multiple breast cancer screening methods in the evaluation of breast non-mass-like lesions,” BMC Med. Imaging 22(1), 202 (2022). [CrossRef]
