Performance of an artificial intelligence tool in percutaneous breast biopsies

Paulina Neira-Vallejos, Departamento de Radiología, Centro de la Mama de Clínica MEDS La Dehesa, Santiago, Chile
María C. Vial, Departamento de Radiología, Centro de la Mama de Clínica MEDS La Dehesa, Santiago, Chile
Carolina Behnke, Departamento de Radiología, Centro de la Mama de Clínica MEDS La Dehesa, Santiago, Chile
Marcelo Poblete-Becker, Departamento de Radiología, Centro de la Mama de Clínica MEDS La Dehesa, Santiago, Chile
M. Rosario van Wersch, Facultad de Medicina, Universidad de Los Andes, Santiago, Chile
Emilia Santelices, Facultad de Medicina, Universidad de Los Andes, Santiago, Chile

Introduction: The use of artificial intelligence (AI) tools in breast ultrasound has shown an increase in the diagnostic accuracy of breast cancer. Objective: To evaluate the diagnostic performance of artificial intelligence (AI) software in the detection of breast cancer in breast lesions biopsied under ultrasound guidance. Material and methods: This prospective diagnostic testing study included the results of patients who underwent ultrasound-guided percutaneous breast biopsy. Images were analyzed and classified by degree of suspicion, first with an ACR BI-RADS category assigned by a specialist radiologist and then by AI software. These were subsequently correlated with the histopathology results obtained from the biopsy, which served as the reference standard. Results: 181 lesions were included in this study, of which 80 were malignant and 101 benign lesions. The sensitivity of the AI tool was 95%, specificity 58%, positive predictive value (PPV) 64%, negative predictive value (NPV) 93%, and accuracy 0.74. The values obtained by the radiologists were 98% sensitivity, 19% specificity, 49% PPV, 95% NPV, and 0.54 accuracy. Conclusions: The AI software demonstrated superior specificity and PPV compared to radiologists in lesions biopsied and could therefore be used to reduce the number of biopsies with benign results.

Keywords: Ultrasound. Breast. Core biopsy. Artificial intelligence.