Multimodal AI Model Accurately Classifies BI-RADS 4a Breast Nodules, Aims to Reduce Unnecessary Biopsies

Multimodal AI model combining ultrasound radiomics, hemodynamics, and clinical data achieved AUC 0.95 (95% CI 0.91-0.99) and 100% sensitivity for classifying BI-RADS 4a breast nodules, potentially avoiding 32.4% of unnecessary biopsies.

• Prospective, multicenter study with temporally independent validation set (training n=260, test n=90).
• Specificity was 86.1% (95% CI 75.8%-93.1%), significantly outperforming logistic regression, SVM, and random forest (p<0.05).
• SHAP analysis identified wavelet texture features, penetrating vessels, and BRCA mutation status as top predictors.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.