Evaluation of Carebot AI MMG Medical Device for Breast Lesion Detection and Density Assessment

A multicenter, multi-reader, retrospective study was designed to validate the clinical efficacy of the proposed Carebot AI MMG (also referred to as "DLAD"). Using a non-certified medical device, a test set of retrospectively collected mammography studies in standard projections (CC and MLO). The performance of the DLAD was evaluated against the ground truth for individual indications (breast density evaluation, breast lesion detection) using Accuracy.

Carebot AI MMG v2.2 classified mammography studies according to the ACR BI-RADS 5th edition, i.e. breast tissue density assessment into A/B/C/D classes. The performance of the Carebot AI MMG device was assessed relative to the ground truth and then compared with the performance of five independent radiologists with varying levels of experience (RAD 1-RAD 5). A rigorous statistical analysis was used in the BI-RADS breast density classification to evaluate the performance of each method - the proposed Carebot AI MMG v2.2 medical device and the compared radiologists in the multi-reader study. The analysis focused on key metrics including Accuracy, F1 Score (Macro-Averaged), Precision (Macro-Averaged), Recall (Macro-Averaged) and Cohen's Kappa (κ) to assess the strength of agreement. Given that all scans were evaluated by all radiologists in the comparison, a bootstrapping method that involves resampling the test data 1000 times.

The medical device (DLAD, Carebot AI MMG v2.2) analyzed mammography studies in standard projections (CC and MLO) and classified the presence of lesions ("present" x "absent") at the mammography study level. DLAD performance was assessed relative to the ground truth and then compared to the performance of five independent radiologists with varying levels of experience (RAD 1-RAD 5). The investigators quantified the performance of diagnostic tests based on Sensitivity, Specificity and Balanced Accuracy. The investigators further assessed the statistical significance of differences between DLAD and individual radiologists using appropriate statistical tests. To determine the reliability of the metrics examined, the investigators calculated 95% confidence intervals using Wilson scores. To evaluate the statistical significance of differences in Sensitivity and Specificity between DLAD and individual radiologists, the investigators applied McNemar's test with a continuity correction.