In this study, we retrospectively analyzed characteristics of small BI-RADS 4A breast lesions, which featured non-parallel on US. The results showed that menopausal status and margin were independently associated with malignancy. A predictive model was then built and presented as a nomogram. The AUC of ROC was 0.747 in the training cohort and 0.741 in the test cohort. Finally, we stratified the breast lesions into a low-risk group, medium-risk group, and high-risk group according to the malignancy risk calculated by nomogram. 10% of patients had malignant pathology in the low-risk group in the training cohort, and 4.8% of patients had malignant pathology in the low-risk group in the test cohort.
In BI-RADS lexicon, non-parallel orientation was defined as the longest axis of the breast lesion is not parallel to the skin. Non-parallel orientation was frequently associated with malignant masses and had a high predictive value for malignancy. These had been demonstrated in numerous studies. Moreover, some studies have investigated that non-parallel orientation was related to advanced cancer and a poor prognosis15, 20. Significantly, in most previous studies, non-parallel orientation was defined as the anterior-posterior dimension greater than the horizontal dimension. However, round masses and those obliqued to the skin were also considered non-parallel orientation according to the BI-RADS lexicon, which resulted in the difficulties in practical applications and more false-positive cases. Wu et al. quantified the orientation angle in their study and found a significant difference between malignant and benign breast lesions21. Accurate and objective measurement of the orientation angle led to better prediction of malignancy compared to the traditional "taller than wide" orientation method. Similarly, Zhu et al. calculated ALS (angle between the long axis of the lesion and skin) with artificial intelligence and found increased ALS in malignant masses compared with the benign ones22. Nevertheless, there were no significant differences in height-to-width ratio between benign and malignant masses. Our study was the first to investigate the risk factor in non-parallel breast lesions. We retrospectively analyzed the images and the orientation was defined based on subjective judgment but not precise measurement. Some biases were inevitable, although our data were examined by two experienced sonographers. We hypothesized that objective orientation evaluation will improve the prediction of the diagnostic model in the future research.
Small breast lesion remains a diagnostic challenge for conventional ultrasound. The feature of the lesion may not be accurate, especially those less than 1 cm, and it is difficult to analyze according to the Breast Imaging Reporting and Data System23. Studies have proven that the accuracy of breast US in differentiating between benign and malignant masses was lower in small groups. Therefore, many researchers combined traditional ultrasound with elastography or contrast-enhanced ultrasound to improve the diagnostic ability of small breast masses24–26. In our study, elasticity assessment was not significantly different between the benign and malignant lesions. Chen et al. found that margin was the only significant factor in evaluating breast masses ≤ 1cm16. Wu et al. found that spiculated margin and anti-parallel orientation at US were associated with poor prognosis in patients with small breast cancer15. This is generally consistent with our results, that margin was the only statistically significant sonographic feature in multivariate analyses. Ultrasound is more accurate in evaluating the margin of lesions than mammography, especially in populations with dense breasts 27. Consequently, the findings of our study had important implications for clinical practices.
The clinical management of patients with BI-RADS 4A breast lesions has always been a thorny issue. Invasive biopsy can diagnose cancers at an early stage, but it may also increase patients' anxiety and unnecessary costs. Many studies have integrated multiple methods to establish a predicted model for downgrading 4A breast lesions, such as radiomics and artificial intelligence, and they have achieved satisfactory AUC3, 7, 28. Notably, our studies focused on the specific lesions, which featured small size and non-parallel orientation on ultrasound. The nomogram incorporated three variables: age, menopausal status, and margin. The diagnostic efficiency of the model was acceptable, with an AUC of the ROC curve of 0.75. More importantly, the diagnostic model in this study was convenient and widely available in clinical practice. Age was an independent factor related to malignancy, and this has been confirmed in previous studies29. Similarly, age was also an important factor in our study, and as an objective variable, the accuracy of results could be increased. Finally, we further conducted risk stratification. Based on the malignancy ratio in each group, we have different recommendations for patients with different risk degrees. Biopsies are strongly recommended for patients who are in high-risk groups. Whereas intensive follow-up observation can be suggested for patients in low-risk groups. For medium-risk patients, combining other imaging examinations such as MR and contrast-enhanced ultrasound may benefit them more.
There are some limitations in this study. First, this was a retrospective study and only pathologically diagnosed patients were enrolled, but a large number of BI-RADS 4A patients were follow-up observed. This resulted in the selection bias, high malignant ratio and the relatively small size of our sample. In the future research, we will recruit patients with stable imaging examinations in at least one-year follow-up, and conduct some prospective research. Second, our research is a single-center study. The diagnosis of BI-RADS 4A lesions on US is dependent on the institution and sonographers training. Therefore, more data from other institutions should be included to validate this model and increase both its generalizability and clinical applicability.