PurposeEvaluating the efficacy of machine learning for preoperative differentiation between invasive and non-invasive breast cancer through integrated automated breast volume scanning (ABVS) radiomics and virtual touch quantification (VTQ) techniques.MethodsWe conducted an extensive retrospective analysis on a cohort of 171 breast cancer patients, differentiating them into 124 invasive and 47 non-invasive cases. The data was meticulously divided into a training set (n = 119) and a validation set (n = 52), maintaining a 70:30 ratio. Several machine learning models were developed and tested, including Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM). Their performance was evaluated using the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC), and visualized the feature contributions of the optimal model using Shapley Additive Explanations (SHAP).ResultsThrough both univariate and multivariate logistic regression analyses, we identified key independent predictors in differentiating between invasive and non-invasive breast cancer types: coronal plane features, Shear Wave Velocity (SWV), and Radscore. The AUC scores for our machine learning models varied, ranging from 0.625 to 0.880, with the DT model demonstrating a notably high AUC of 0.874 in the validation set.ConclusionOur findings indicate that machine learning models, which integrate ABVS radiomics and VTQ, are significantly effective in preoperatively distinguishing between invasive and non-invasive breast cancer. Particularly, the DT model stood out in the validation set, establishing it as the primary model in our study. This highlights its potential utility in enhancing clinical decision-making processes.
Read full abstract