EP08.34: Complete fetal biometry with 3D ultrasound and image recognition: prospective assessment of reliability and agreement with 2D measurements

To assess the agreement between measurements extracted from 3D US volumes (by applying an image recognition algorithm) and standard 2D US measurements. The secondary objectives were to assess the impact on image quality scoring and reproducibility and the agreement with human-driven prediction issued from the same volume analysis. Datasets were collected from a consecutive sample of women attending US follow-up (singleton 16–30 WG). Each dataset contained 2D measurements for reference (head and abdomen circumference and femoral length) and 3D US volume acquisitions of the fetal head, abdomen and thigh (2 volumes for each target). Two postprocessing procedures were applied to the 3D volumes to obtain measurements and were assessed for completion rates, image quality, agreement with references, and reproducibility. Those two procedures consisted of software-driven and human-driven measurements and relied respectively on algorithm- and operator-based detection of the targeted plans and calliper positioning. Measurements were converted to Z scores and fetal weight estimation for analysis. A total of 175 datasets were collected. Completion rates in achieving measurements ranged between 94% (abdomen, software-driven procedure) and 100% (head, both procedures). A modest weakening in quality (of uncertain clinical significance) was notable for the head and abdomen for the software-driven procedure. Compared to the reference and human driven-procedure, the software-driven procedure tended to slightly overestimate EFW (e.g. 95% confidence interval ranging from 445g to 635g for a 525g-sized fetus at 22 WG). Random error tended to be inflated for fetuses > 700g. Intra- and interoperator reproducibility were appropriate (intraclass correlation coefficient intervals ranged from 0.8 to 0.99). When performed by expert operators, the 3D volume-based complete automation of US biometry in the second trimester seems feasible and presents appropriate reproducibility and image quality scoring.