Performance of an Artificial Intelligence Support System on Screening Mammography Cases Proceeding to Stereotactic Biopsy

Abstract

<b>Background/Objective:</b> The objective was to evaluate the standalone performance of an AI system, Transpara 1.7.1 (ScreenPoint Medical), in screening mammography cases proceeding to stereotactic biopsy using histopathological results as ground truth. <b>Methods:</b> This retrospective study included 202 asymptomatic female patients (mean age: 57.8 years) who underwent stereotactic biopsy at a multicenter academic institution between October 2022 and September 2023 with a preceding screening mammogram within 14 months. Transpara AI risk scores were compared to pathology results (benign versus malignant). Performance metrics for AI including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristic curve (AUC) were calculated. <b>Results:</b> Transpara AI classified 20 of 39 malignant findings (51%) as elevated risk compared with 50 of 211 total findings (24%). AI score was positively correlated with malignancy (r = 0.29, <i>p</i> < 0.001). Sensitivity for detecting malignancy (classifying as intermediate or elevated risk) was 94.9% (95% CI: 81.4-94.1), specificity was 24.4% (95% CI: 18.3-31.7), PPV was 22.2% (95% CI: 16.3-29.4), and NPV was 95.5% (95% CI: 83.3-99.2). Transpara had fair performance in detecting breast cancer with AUC 0.73 (95% CI: 0.63-0.82). <b>Conclusions:</b> Transpara AI is a useful screening mammography triage tool. Given its high sensitivity and high negative predictive value, AI may be used to guide radiologists in making biopsy or follow up recommendations. However, the high false-positive rate and presence of two false negatives underscore the need for radiologists to use caution and clinical expertise when interpreting AI results.

Ähnliche Arbeiten