Objective To screen the ultrasound features of atypical invasive ductal carcinoma and to establish Logistic regression prediction models to assist in the diagnosis.Methods The preoperative ultrasound images of 42 cases diagnosed by ultrasound as BI-RADS 4A lesions yet confirmed by pathology as invasive ductal carcinoma, were retrospectively analyzed. Logistic regression models were established, and the receiver operating characteristic curve (ROC) was applied to evaluate the diagnostic efficacy of the models.Results The results of multivariate Logistic regression analysis showed that unclear tumor margin [O^R = 23.371 (95% CI: 2.207, 247.442)], microcalcification [O^R = 5.120 (95% CI: 1.481, 17.697)] and resistance index (RI) > 0.7 [O^R = 12.912 (95% CI: 2.579, 46.165)] were of diagnostic significance for atypical invasive ductal carcinoma. The regression equation of the diagnostic model based on these indicators for atypical invasive ductal carcinoma was Logistics (P) = -1.674 + 3.152 × X1 (microcalcification) + 1.633 × X2 (unclear tumor margin) + 2.390 × X3 (RI > 0.7). When the cut-off value of the diagnostic model was set as 0.380, the area under the ROC curve (AUC) was 0.804 (95% CI: 0.711, 0.897), with a sensitivity of 85.7% (95% CI: 71.5%, 95.6%), a specificity of 59.1% (95% CI: 43.2%, 73.3%), a Youden index of 0.448, and an accuracy of 72.1%.Conclusions The Logistic regression model based on ultrasound signs exhibits a high sensitivity and clinical applicability for the diagnosis of atypical invasive ductal carcinoma.