Abstract
Background The most common cancer in the world, breast cancer (BC), poses serious problems to healthcare. Making an accurate diagnosis of these patients' HER2 status is essential for therapy planning.Methods A prospective cohort of patients with BC was enrolled between June 2020 and october 2023. The patient's clinical data and features from their ultrasonography were gathered. Postoperative tumor pathology specimens were subjected to immunohistochemistry and fluorescence in situ hybridization examinations to ascertain the HER2 status. Lasso regression was used to choose characteristic variables. Univariate and multivariate logistic regression analysis were used to find the HER2 status-independent factors. The performance of the nomogram model was then assessed using calibration curves and decision curve analysis (DCA).Result 97 (22.25%) of the 436 BC patients enrolled in the research had positive HER2 results. Progesterone receptor expression, Ki-67 levels, and estrogen receptor expression differed statistically amongst patients with different HER2 statuses. Lasso regression identified six ultrasonographic variables closely associated with HER2 status from a pool of 786 features, leading to the generation of a radiomic score for each patient. Multivariate logistic regression analysis revealed that PR (OR = 0.15, 95%CI = 0.06–0.36, p < 0.001), Ki-67 (OR = 1.02, 95%CI = 1.00-1.03, p = 0.012), and Radiomic score (OR = 5.89, 95%CI = 2.58–13.45, p < 0.001) were independent predictors of HER2 status. The nomogram model demonstrated areas under the curve (AUC) of 0.823 (95% CI = 0.772–0.874) and 0.812 (95% CI = 0.717–0.906) in the training and validation cohort, respectively.Conclusions A methodology that integrates clinical data, cutting-edge imaging, and machine learning to provide individualized treatment plans is presented for the non-invasive prediction of HER2 status in breast cancer.