Abstract
Existing artificial intelligence (AI) breast ultrasound solutions have limitations owing to their non-real-time detection and server dependency. However, novel real-time AI solutions enable on-device detection and differential diagnosis, aiding immediate decision-making. This study evaluated the feasibility of real-time artificial intelligence-based computer-aided detection/diagnosis (AI-CAD) for breast ultrasound in a clinical setting and assessed its preliminary efficacy in comparison with expert evaluations. A feasibility study was conducted from August to December 2023 at a tertiary medical center in Taiwan using a real-time AI solution (CadAI-B for Breast cancer). AI-CAD runs on a tablet PC and streams the display output from the ultrasound vendor's device via HDMI or DVI. Real-time AI-CAD was evaluated for detection and diagnostic performance based on sensitivity, specificity, and area under the curve (AUC). The analysis included 33 patients with 14 malignancies, 17 benign lesions, and 2 normal cases; 30 (90.9%) underwent biopsy. AI-CAD successfully identified all malignancies in real-time. As AUCs were calculated using the malignancy score and Breast Imaging Reporting and Data System (BI-RADS), the overall diagnostic performances were 0.835 and 0.850, respectively. The per-patient sensitivity and specificity were 100.0% and 52.6%, respectively. The BI-RADS distribution was the same between AI-CAD and experts in malignant cases. In benign cases, AI-CAD categorized nine (50.0%) as C4A or C4B, whereas experts classified 13 (72.2%), indicating the potential to reduce the need for biopsy. Real-time AI-CAD is feasible for supporting detection during breast ultrasound scanning, with potential efficacy in aiding differential diagnosis and reducing the risk of unnecessary biopsies.