EP01.59: AI‐driven follicle measurement: evaluation of SonoAVCTM follicle auto caliper

Assessing the number and size of follicles is an essential step during fertility treatments. Measuring follicular size during stimulation is of particular importance as it provides an indication of oocyte maturity. However, such measurements are time-consuming and prone to intra- and interobserver variability, therefore warranting the development of automated techniques. SonoAVC™ follicle Auto Caliper (GE HealthCare, Zipf, Austria) is a novel semi-automated tool that places two-dimensional (2D) calipers on follicles based on user interaction, namely tap or click. This study is aimed to assess its performance on retrospective data. The semi-automatic caliper placement was developed based on the segmentation outputs of a deep neural network. The tool places 2 orthogonal calipers onto user-selected follicles. Ground truth (GT) annotations for training and validation were contoured by junior medical professionals and reviewed by a senior obstetrics and gynecology practitioner. The offline evaluation of the SonoAVC™ follicle Auto Caliper tool was performed in an inhouse developed tool on 14 volumes. Manual 2D calipers were placed by a senior and junior medical staff. Follicle count measurements were performed during postprocessing. Bland-Altman plot analysis revealed smaller limits of agreement between the calipers placed on the automatic predictions and GT annotations on operator selected slice (senior: 0.92 and –1.22; and novice: 1.19 and –1.35) than between the senior and GT (1.35 and –1.24); and junior and GT (1.34 and –1.52). The tool's performance is comparable to the operators in terms of follicle counts. The evaluation indicates that users employing SonoAVC™ follicle Auto Caliper reached a better or similar performance, than users placing the calipers manually. SonoAVC™ follicle Auto Caliper accelerates the clinical workflow by reducing the keystroke amount from four clicks/taps to one.