Skip to main content
. Author manuscript; available in PMC: 2023 Mar 2.
Published in final edited form as: NEJM Evid. 2022 Mar 28;1(5):10.1056/evidoa2100058. doi: 10.1056/evidoa2100058

Table 2.

Gestational Age Estimation of Deep-Learning Model Compared with Trained Sonographer in the Main Test Set and In Vitro Fertilization (IVF) Test Set.*

Main Test Set (n = 716)
IVF Test Set (n=47)
Estimate Model Biometry Difference (95% CI) Model Biometry Difference (95% CI)
MAE (±SE) — d   3.9±0.12   4.7±0.15 −0.8 (−1.1 to −0.5)   2.8±0.28   3.6±0.53 −0.8 (−1.7 to 0.2)

RMSE (±SE) — d   5.1±0.18   6.1±0.19 −1.1 (−1.5 to −0.6)   3.4±0.31   5.1±0.67 −1.7 (−2.8 to −0.5)

Estimate of bias (±SE) — d§ −0.7±0.19   1.3±0.22   1.3±0.46   1.7±0.70

First trimester — d

 MAE (±SE)   2.1±0.19   2.0±0.23   0.2 (−0.4 to 0.7)   2.3±0.34   2.2±0.38

 Estimate of bias (±SE)   0.8±0.32   1.4±0.29   0.8±0.67   0.9±0.66

Second trimester — d

 MAE (±SE)   3.1±0.16   3.3±0.17 −0.2 (−0.5 to 0.1)   2.7±0.45   3.4±0.84

 Estimate of bias (±SE)   0.2±0.25   0.7±0.26   1.3±0.69   2.0±1.04

Third trimester — d

 MAE (±SE)   4.7±0.18   6.0±0.22 −1.3 (−1.8 to −0.8)   3.7±0.69   5.7±1.38

 Estimate of bias (±SE) −1.4±0.29   1.6±0.36   1.8±1.23   2.4±2.13

Absolute error (±SE) — %

 <7 d 86.0±1.3 77.0±1.6   9.1 (5.7 to 12.5) 95.7±2.9 83.0±5.5

 <14 d 98.9±0.4 96.9±0.6   2.0 (0.5 to 3.4) 100.0 100.0

North Carolina — d

 MAE (±SE)   3.4±0.15   4.0±0.20 −0.6 (−0.9 to −0.2)

 Estimate of bias (±SE) −0.3±0.23   0.7±0.28

Zambia — d

 MAE (±SE)   4.4±0.19   5.4±0.22 −1.0 (−1.5 to −0.5)

 Estimate of bias (±SE) −1.0±0.30   1.9±0.35
*

CI denotes confidence interval, MAE mean absolute error, and RMSE root mean square error

The main test set comprises a 30% random sample of participants who are dated by a prior ultrasound and who are not included in the IVF or novice test sets. Participants were enrolled in either North Carolina or Zambia. Blind sweeps and fetal biometry were collected by a sonographer on a commercial ultrasound machine

The IVF test set comprises studies conducted in women who conceived by in vitro fertilization. All participants were enrolled in North Carolina. Blind sweeps and fetal biometry were collected by a sonographer on a commercial ultrasound machine.

§

Estimate of bias is reported as the estimated mean of the error.

Trimesters are defined as 97 days or less, 98 to 195 days, or 196 days or more.