Preoperative ultrasound radiomics analysis for expression of multiple molecular biomarkers in mass type of breast ductal carcinoma in situ

Breast ductal carcinoma in situ (DCIS) is a kind of malignant tumor originated in the ductal epithelial tissue, limited to the basement membrane [1]. DCIS is the second most common breast tumor, and accounts for approximately 20–30% [2]. Some DCIS had the potential to further develop into breast invasive cancer [3]. The clinical treatments of patients with DCIS include surgical resection, radiotherapy, chemotherapy and endocrine therapy, in which surgical resection includes simple focal resection and mastectomy, with different therapeutic effects [4]. Although the prognosis of DCIS is good, more than 14% of DCIS patients may develop invasive cancer without treatment within 10 years [5]. In the past 10 years, the incidence of DCIS has gradually increased, highlighting the understanding the importance of DCIS pathology [6]. However, the pathologic mechanism of the transition from DCIS to invasive carcinoma is still unclear, which produces clinical challenges of overdiagnosis and overtreatment in patients with DCIS [7]. Therefore, the investigators thought more studies were need to understand the potential of the pathological process of DCIS, in order to adapt to the current individualized, refined treatment.

Immunohistochemistry (IHC) can reflect the expression of molecular biomarkers in tumor tissue, which can further clarify the biological behaviors of tumors. The expression of different molecular biomarkers can lead to different biological behaviors and treatments. Some studies have shown that some molecular biomarkers were important indicators for predicting biological behavior and judging follow-up treatment in patients with DCIS, such as estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), Ki67, p16, and p53. ER and PR are the earliest molecular biomarkers of breast cancer. They are predictors of breast cancer prognosis and endocrine adjuvant therapy [8]. HER2 is a proto-oncogene, which is mainly involved in tumor signal transduction and cell proliferation. Its positive expression can lead to a high distant metastasis rate and poor prognosis of breast cancer. Ki67 is an antigenic nuclear protein that can be used as a proliferation marker. Its high expression is considered to be a biomarker of tumor invasion [9]. Ki67 has a good application prospect in predicting endocrine therapy response of breast cancer [10]. Defined as a tumor suppressor gene, p16 is considered to be an important cell cycle regulator [11]. P16 is closely related to abnormal methylation initiation. P53 is a common tumor suppressor gene. Impaired function of p53, such as p53 mutation, can lead to uncontrolled proliferation of damaged cells [12]. Therefore, accurate identification of the expression of molecular biomarkers can help stratify tumor risk and facilitate the development of personalized and accurate treatment plans.

Currently, the preoperative evaluation of the molecular biomarkers of DCIS mainly depends on IHC detection after biopsy. However, because the progression of tumors are dynamic process, there are differences in spatio-temporal evolution. In addition, the evaluation results of a few tissue biopsies do not necessarily represent the expression of the molecular biomarkers of the whole tumor [13]. Invasive procedures and potential risks limit its multiple applications in monitoring tumor progression and biological behavior. However, the preoperative monitoring of molecular biomarkers can dynamically identify the progression of tumors and the changes in biological behavior, which has great significance for the accurate formulation of treatment plans and the evaluation of curative effects. To avoid overdiagnosis and overtreatment of patients with DCIS, it is necessary to provide dynamic and accurate evaluation of biological behavior information for the clinic.

With the breakthrough of imaging technology, mammography is an important examination method for DCIS, which is sensitive to the detection of calcification [14]. Ultrasound (US) has become the main examination technology to detecting breast lesions [15], which is real-time, dynamic and non-invasive. There are two types of DCIS: mass and non-mass. Some studies suggested that the detection rate of US in 93 patients with mass type DCIS reached 77.4% [16]. The main characteristics of DCIS in ultrasound were: uneven low or slightly low echo, irregular shape, unclear borders, parallel skin, weakened posterior echo, calcification, and some blood flow signals [17, 18]. However, mammography or US-assisted screening could increase overdiagnosis because both tests primarily detect low-grade invasive cancers [19]. At the same time, radiologists are very time-consuming to accumulate experience and have strong personal subjectivity, which is another problem that needs to be solved. There is an urgent need for more advanced imaging evaluation methods to guide the diagnosis and treatment of DCIS.

Breast lesions are diagnosed and screened by various imaging methods, such as mammography, US, and magnetic resonance imaging (MRI). All three examination have some limitations [20]. Radiomics is a hot subject of artificial intelligence that is applied in the medical imaging field, which is the cornerstone of precision science in the future. Radiomics is defined as the extraction of high-throughput features from single or multiple medical image patterns to select features that are closely associated with tumors,and the ultimate goal is to construct prediction models based on features to provide accurate tumor phenotypic analysis information and accurate treatment decision-making [21]. Radiomics highlights the image features that are not visible to the naked eye, thus significantly enhancing the predictive power of medical imaging [22]. Radiomics has been developed in a wide range of fields, such as disease diagnosis and biological behavior judgment. For example, the US-radiomics model developed by Luo WQ et al. had better performance in distinguishing breast lesions than breast imaging reporting and data system (BI-RADS) [23]. Lin F et al. found that the radiomics score was more effective than the clinical radiological model in benign and malignant breast lesions (< 1 cm) [24]. These series of studies showed that radiomics had better performance than traditional imaging features in the diagnosis of breast diseases to some extent.

Thus, this retrospective study intended to further clarify the relationship between US-radiomics and molecular markers of DCIS. Radiomics models had been developed to noninvasively evaluate the expression of molecular markers to help achieve accurate risk stratification and treatment for patients with DCIS.

This study was the first non-invasive comprehensive analysis based on US-radiomics to predict the expression of molecular markers of DCIS. The investigators recruited only 116 patients with DCIS for this study, but it was exciting to see that the radiomics models showed more than moderate predictive performance in predicting molecular biomarker expression of DCIS.

DCIS is a malignant tumor with good prognosis, but it is heterogeneous in morphology and genetics. Before the imaging examination was performed, the diagnosis of DCIS was only due to the appearance of nipple discharge and/or palpable mass symptoms, which accounted for only 2% of DCIS detected. It showed that DCIS with hidden symptoms were easily missed [35]. With the screening of imaging technology (mammography, US and MRI), the detection rate of DCIS had gradually increased. This detection rate included symptomatic DCIS, and whether there was overdiagnosis in the detection of insidious DCIS was also a hot topic of controversy [36], Unfortunately, the diagnosis of DCIS marked women as at risk of invasive breast cancer, so women diagnosed with DCIS may suffer serious psychological distress, leading to the progression of DCIS [37]. In addition, the current treatment methods were also facing the controversy over the treatment of some patients [38]. Therefore, the main clinical challenge in DCIS has been to distinguish between patients who have a better chance of developing invasive cancer and require more treatment and those who are less likely to develop DCIS and need less or no treatment [39]. Immunohistochemical markers can explain the changes in the biological behaviors of tumors on the molecular level. More and more studies have pointed out the changes in molecular markers associated with the progression of DCIS to invasive cancer [40]. For example, Zhang GJ et al. [41] found that 79% of DCIS patients were positive for P53 when studying the occurrence and development of breast cancer. Davis et al. [42] demonstrated that high Ki67 expression was an independent predictor of postoperative recurrence in patients with DCIS. Cornfield DB had found a higher recurrence rate with PR > 3.5% using tree structure survival [43]. The results showed that the changes in the biological behavior of DCIS were closely related to the expression of molecular biomarkers.

About various imaging technologies, they also have application limitations [44]. Mammography is the main method of early breast cancer detection, but it is closely related to the density of the lesion and the possibility of covering the lesion [45]. However, Chinese women have dense breasts, so they had certain limitations in finding suspicious lesions in the dense tissues of breasts through mammography [46, 47]. The traditional mammography diagnosis method will cause trauma to the patient to a certain extent and reduce the patient's treatment compliance. Due to its high sensitivity to soft tissues, US can better show lesions in dense glands and has become the primary imaging method for Chinese women to screen and diagnose breast diseases. For non-mass DCIS, US is difficult to recognize [48]. Therefore, this retrospective study only examined mass DCIS, which is a limitation of the study. MRI has considerable advantages in detecting breast lesions, but its specificity is limited by several factors that affect image quality, such as magnetic field and gradient strength, coil performance, contrast agent efficacy and menstrual cycle [49].

Radiomics mainly studies the quantitative features that are related to biology in medical images. Radiomics features are considered the invisible tissue infrastructure components of the object to be imaged, which can serve as a valuable method for studying cancer by imaging, such as MRI. Radiomics can provide in vivo visualization and quantitative analysis of the imaging features of the whole imaging mass. Therefore, radiomics is a precision medical method for non-invasive diagnosis, evaluation of efficacy, biological behavior [50]. Currently, radiomics mainly relies machine learning algorithms to identify meaningful features of image training data set, and for further interpretation of the information and the optimization, so as to accurately predict the content of the research. An independent data set is applied to test the universality of the model and provides feedback for further optimization of the model [51]. To a certain extent, it improves the utilization of image information and enables differential diagnosis of diseases on more subtle levels that cannot be recognized by the naked eye.

Breast radiomics studies are mostly applied to the prediction of the molecular classification, lymph node metastasis and molecular markers of invasive ductal carcinoma. For example, Demircioglu A et al. [52] constructed radiomics models for predicting Ki67 expression in invasive breast cancer based on eight features extracted from MRI images, with an AUC of 0.81. Zhou et al. [53] explored the significance of MRI-radiomics models for predicting the expression of HER2 in patients with invasive breast cancer before surgery; the validation set AUC reached 0.81. There are few reports on DICS with radiomics. However, there are clinical challenges in the diagnosis and treatment of patients with DCIS. Tumor progression and treatment decisions are affected by multiple tumor molecular biomarkers, which require to comprehensively analyze and evaluate the molecular biomarkers of DCIS. To expand the application of radiomics in DCIS, the investigators carried out this study to assess the feasibility of molecular biomarkers of DCIS. The investigators believe that information obtained from multiple molecular biomarkers can help explain the underlying pathological process of DCIS.

In this study, the first highlight, the first comprehensive analysis of molecular markers of DCIS was conducted based on radiomics. Second highlight, there were thousands of radiomics features, including eight classifications: original feature can reflect the number of voxels in the images, intensity distribution, pixel pair frequency, image average gray value, size and shape of ROI (https://​pyradiomics.​readthedocs.​io/​en/​latest/​features.​html); Ipris features capture nodular heterogeneity and differential growth patterns; CoLIAGe features can distinguish disease phenotypes that have similar morphologic appearances [54]; wavelets features represent most of the edge information in images; Shearlet features are better for processing high-dimensional signals; Gabors features extract the edge and gradient information of image and reflect the spatial frequency feature; PLBP, and WILBP: PLBP features are an oriented local texture descriptor that combines the phase congruency approach with the LBP. The third highlight of this study was to construct dozens of prediction models by combining multiple classifiers with multiple feature selection to select the optimal prediction results. RF and SVM-RFE had significant performance in feature selection of multiple molecular markers, KNN and SVM classification performed well too. Finally, through the verification of the test set, the prediction models all showed moderate performance.

There were also some shortcomings in our study. First, this retrospective study had the problem of small sample size. It was necessary to increase the sample size or multi-center cooperation to construct universal models. Second, when the radiologists manually delineated the ROIs, there were a certain degree of subjectivity to the contours of the lesions, which may lead to poor robustness of the models. In addition, the delineation process was done by only one radiologist. Third, the investigators only investigated the features extracted from the largest section, which could not represent the whole tumor. Due to the limitations of US, it was not possible to conduct three-dimensional studies similar to other imaging studies.

The application of machine learning-based radiomics analysis provided a non-invasive method for predicting the expression of multiple molecular biomarkers in DCIS, with good prediction performance. This study also demonstrated the potential of radiomics in pathologic assessment and individualized precision therapy.