Artificial intelligence based on automated breast volume scanner (ABVS) in breast nodules:A deep learning model for diagnosis

Objectives: We aimed to explore the use of artificial intelligence (AI) in differential diagnosis of benign and malignant breast nodules with automated breast volume scanner (ABVS). Methods: The AI model used in this study is based on the VNet, trained with a collection of breast nodules. Another collection of breast nodules were used to test the diagnostic accuracy among AI model, radiologists with different experiences, and AI-assisted junior radiologist. Sensitivity, specificity, accuracy, positive predictive value, negative predictive value and area under the curve (AUC) were used as indices to evaluate the efficiency on distinguishing between benign and malignant breast nodules of four different diagnostic groups. Results: The training cohort includes 100 breast nodules, the validation cohort includes 47 breast nodules. Among these 147 cases, 92 cases were benign (62.6%), 55 cases were malignant (37.4%). The AUC values of AI diagnosis, independent diagnosis by senior radiologist, independent diagnosis by junior radiologist, and AI-assisted diagnosis by junior radiologist were 0.839(95%CI, 0.703-0.930), 0.956(95%CI, 0.853-0.994), 0.814(95%CI, 0.674-0.912), and 0.963(95%CI, 0.863-0.996), respectively. The diagnostic efficiency of junior radiologist improved with the assistance of AI (AUC = 0.814 vs 0.963, Z=2.772, p 0.05). Conclusions: The AI model could distinguish between benign and malignant breast nodules with ABVS. AI model is the fastest diagnosis method with equivalent diagnosis efficiency with experienced radiologists, and could improve the diagnostic ability of junior radiologist.