In this study, multimodal ultrasound features including linear or segmental distribution, posterior shadowing, calcification, vascularity, high elasticity scores of 4 and 5 were identified as significant factors associated with malignant breast NMLs. The nomogram’s performance utilizing these features was found to be comparable to that of experienced radiologists, and exhibiting superior specificity and PPV. Furthermore, the nomogram displayed good calibration and satisfactory clinical utility both in the training and validation cohorts.
Since the introduction of breast NMLs on ultrasound, several studies have presented findings on the utilization of various ultrasound techniques either independently or in conjunction for assessing NMLs [23–28]. These studies have demonstrated varying degrees of enhancement in diagnostic accuracy and specificity of biopsy decisions. Nevertheless, a consistent characterization and classification system for NMLs on ultrasound have been lacking in these studies, potentially resulting in discrepancies in interpretation and assessment among different radiologists. For instance, Ko et al. categorized NMLs into four types and assigned BI-RADS categories based on their PPVs, thereby establishing a dependable reference for stratifying NML risk [13]. However, this system exhibited a higher malignancy incidence (10%-79%), which could lead to an increase in unnecessary biopsies. Another classification system, developed by Park et al., based on suspicious ultrasound findings or in combination with mammographic features, aimed to aid in the interpretation and management of breast NMLs [11]. The results indicated that this system significantly improved diagnostic performance among radiologists, with an AUC ranging from 0.951 to 0.956, and specificity increasing from 49.3–76.8%. Choi et al. further outlined a standardized interpretation algorithm flowchart for BI-RADS classification of NMLs using ultrasound features, this approach has demonstrated a reduction in false-positive rates on NML management in their clinical practice [11,16,17,29]. However, the efficacy of this approach requires further validation across diverse institutions and scenarios. Recently, nomograms have emerged as valuable tools for establishing intuitive relationships between evaluation variables, offering quantitative and personalized methods for predicting cancer risk [30–32]. In our study, we have presented a quantitative association between multimodal ultrasound features and malignant breast NMLs, with the nomogram based on these features exhibiting satisfactory performance.
In a prior study involving 229 cases, a nomogram was developed using patient age, clinical symptoms, and ultrasound features to predict malignant NMLs in the Asian population, demonstrating favorable diagnostic accuracy and clinical utility [33]. Of which, the meaningful ultrasound features encompass orientation, echo patterns, calcification, and vascularity graded by Adler’s classification. But in our study, we only focused on the ultrasound features for the following reasons: ① the contentious relationship between breast cancer and age [34,35], ② the majority of the participants in this research exhibit clinical manifestations, and this ultrasound examination is mostly used for diagnostic purposes, ③ multiple ultrasound descriptions of breast NMLs have been documented in existing literature, warranting further investigation and discussion. The current study incorporated proposed features and descriptors of NMLs, along with a larger sample size. Ultimately, our results exhibited higher AUC, sensitivity, and specificity compared to previous studies, and comparable to the performance achieved by experienced radiologists using BI-RADS categories. It is worth mentioning that it is important to acknowledge the considerable dependence of radiologists on their experience when characterizing and assessing breast NMLs.
In order to enhance the utility of multimodal ultrasound features in clinical practise, a nomogram was developed. This involved assigning numerical values to each individual ultrasound feature, with the total points for each NML corresponding to a specific malignant risk value for use in subsequent clinical decision-making. Of which, the elasticity score emerged as the most influential feature in predicting malignancy in breast NMLs. The strain elasticity score demonstrated a notable specificity of 93.8% for evaluating NMLs when the threshold was set between 3 and 4, as reported in a previous study [24]. In our study, a score of 4 and 5 exhibited the highest ORs (7.00, 15.77) and points (70, 100). The linear and segmental distribution patterns detected on ultrasound were found to be indicative of lesions within ducts and branches, often suggestive of ductal carcinoma in situ (DCIS) or suspicious multifocal breast cancer [36]. Compared to a previous study reporting an OR of 3.65 [11], the current study revealed higher ORs of 4.69 and 7.67 for linear and segmental distributions, with corresponding nomogram point values of 56 and 74. Additionally, the presence of calcifications on ultrasound was identified as a significant risk factor for breast cancer, which have been reported to be more than three times more likely to be malignant [5,19,37]. This was associated with an OR of 7.40 and assigned 72 points in predicting malignant NMLs in the present study. The presence of posterior shadowing on ultrasound imaging may indicate pathological alterations that stimulate the proliferation of connective tissue, leading to attenuation of the ultrasound beam [38]. This phenomenon can be observed in both benign and malignant lesions. Our findings revealed an OR of 3.14 and 42 points for malignant NMLs. Internal vascularity within a focal isoechoic or hypoechoic area was noted to aid in identifying NMLs, with PPVs of hypervascular for malignant NMLs ranging from 27.5–90.5% [19]. In our research, over half (57.2%, 328/573) of the NMLs were found to associated with hypervascular, and 67.7% of these were subsequently verified as malignant. Architectural distortion and duct changes are common features in breast NMLs, and more frequent in malignant lesions [1,10,39,40]. These findings indicated a statistically significant association with malignancy in the univariate logistic regression analysis, however, this association was not observed in the multivariate logistic regression analysis. Considering the potential overlap of features between benign and malignant lesions in NMLs, pathological findings in this study may offer insights into this scenario. It is noteworthy that benign lesions such as adenosis, mastitis, and intraductal papillomas have been significantly identified in NML cases exhibiting structural distortion or ductal changes. Multicenter studies are warranted to be carried out in the future.
The present study had some limitations. Firstly, it was a retrospective, single center study with solely internal validation of the retrospective data, further investigation is necessary to assess the predictive precision of the nomogram for ultrasound NMLs and its practical application through additional multi-center external validation and prospective research studies. Secondly, all cases included in this study were confirmed by biopsy or surgical pathological results, potentially introducing bias in patient selection. Lastly, the institution where the study was conducted serves as a breast tumor diagnosis center with a substantial number of referral patients, utilizing ultrasound examinations primarily for diagnostic purposes. Given the relatively high proportion of malignant breast NMLs in this setting, the findings may not be directly generalizable to screen all populations.
