A Simplified Approach to Predicting the Malignancy of Breast Lesions: Nomogram in Ultrasonography

Data Collection: This study retrospectively collected clinical and ultrasound examination data from patients who underwent breast lesion surgery at our hospital from January 2020 to June 2023. Inclusion criteria included patients with complete clinical information and available ultrasound image data. Parameters extracted from this data included age, 2D ultrasound images, Doppler ultrasound images, and ultrasound diagnostic reports. Feature extraction from ultrasound images included 2D lesion information (maximum diameter, orientation, echogenicity, morphology, margins, calcification type, ductal changes), Doppler information (blood flow pattern, resistance index), and BI-RADS classification based on suspicious ultrasound findings by physicians.

Model Development: Firstly, we conducted multicollinearity analysis using Variance Inflation Factor (VIF) to select variables with VIF less than 5, aiming to reduce the impact of collinearity. We used post-operative pathological results of breast lesions as the gold standard for model development. In the R programming language, we utilized the caret package to randomly split the final samples into training and validation sets in a 7:3 ratio based on the outcome variable (benign or malignant breast lesions) while setting a random seed (set.seed) for result reproducibility. Subsequently, we performed univariate logistic regression analysis on binary variables in the training set, retaining variables with P < 0.05, followed by multivariate logistic regression analysis to identify independent predictors of breast lesion malignancy.

Model Validation: To validate the model's performance, we constructed a nomogram based on the weight allocation of each independent predictor. Then, we comprehensively validated the model in the validation set, including calculating sensitivity, specificity, accuracy, and concordance. Receiver Operating Characteristic (ROC) curves were plotted, and the area under the curve (AUC) was calculated to determine the optimal threshold for quantitatively predicting the probability of breast cancer occurrence in patients. Additionally, we performed Decision Curve Analysis (DCA) to assess the net clinical benefit of the model at different patient decision thresholds. DCA helps determine the practical utility of the model in clinical decision-making and identifies the optimal threshold for predicting the probability of disease occurrence, aiding physicians in making better decisions. These validation metrics were used to evaluate the model's performance, accuracy, and potential application in real clinical practice.