Repeatability and reproducibility of quantitative ultrasound cervical measurements in women at risk for preterm birth

Abstract

Purpose: Transvaginal Quantitative Ultrasound (QUS) has the potential to enhance preterm birth risk monitoring during pregnancy but evidence on the reliability of commonly used QUS parameters in vivo is lacking. This study assesses intra-sonographer repeatability and inter-sonographer reproducibility of six different QUS measurements of the human cervix during pregnancy: Attenuation Coefficient (AC), Lizzi Feleppa (LF) Intercept, Midband and Slope, and Envelope Kappa and Mu.

Methods: This prospective study was approved by the institutional review board at the University of Illinois, Chicago. Informed consent was obtained from all participants, who were selected from pregnant women enrolled in the single-center study, "QUS Technology for Identifying At-Risk Women for Spontaneous Preterm Birth." They received a standard clinical transvaginal ultrasound scan followed by two research scans at 20 ± 2 and 24 ± 2 weeks of gestation. During one or both research scans, they underwent two independent examinations (same-sonographer or cross-sonographer). QUS measurements were computed from ultrasound radiofrequency (RF) data. Variation attributable to transducers, phantoms, and sonographers was evaluated by linear mixed model analysis. QUS parameters were averaged over four acquisitions from each examination. Repeatability and reproducibility were evaluated using the coefficient of variation (CoV), intraclass correlation coefficient (ICC), and Bland-Altman analysis.

Results: Eighty-one participants (mean age, 27.12 years ± 5.81) were recruited for a total of 82 examination pairs, yielding 36 intra-sonographer and 46 inter-sonographer pairs of examinations. Transducer and reference phantom variances were not statistically significant (p > 0.05). AC, LF Midband, LF Slope, Kappa, and Mu displayed moderate re-examination (intra-sonographer) repeatability (CoV: 11.9%-12.9%, ICC: 0.62-0.69). LF Intercept had poor repeatability (CoV: 6.9%, ICC: 0.38). AC and LF Midband also displayed moderate inter-sonographer reproducibility (CoV: 10.4%-13.9%, ICC: 0.61-0.63). LF Intercept had marginal reproducibility (CoV: 7.6%, ICC: 0.51). Kappa and mu had poor reproducibility (CoV: 6.9%-14.5%, ICC 0.27-0.38).

Conclusion: Averaged in vivo transvaginal QUS measurements of AC and LF Midband have potential for development as noninvasive clinical measurements with moderate reproducibility for auxiliary monitoring of the progress of pregnancy. The other QUS parameters evaluated in this study require further refinement before they can be recommended for clinical use.

Supplementary information: The online version contains supplementary material available at 10.1007/s44352-025-00020-3.

Keywords: Cervix; Intraclass correlation; Linear mixed model; Preterm birth biomarkers; QUS imaging.

Fig. 1

Participant recruitment flow diagram

Fig. 2

B-mode image of the cervix with FOI and heat map of sub-FOI QUS parameters using Siemens 5–9 MHz vaginal probe, chosen because of its use in clinical practice. Parameters shown are a Attenuation Coefficient (dB/ cm-MHz), b LF Intercept (dB), c LF Midband (dB), d LF Slope (dB/MHz), e Kappa, and f Mu

Fig. 3

Box plots of participant measurement distributions across sonographers for a Attenuation Coefficient (dB/cm-MHz), b LF Intercept (dB), c LF Midband (dB), d LF Slope (dB/MHz), e Kappa, and f Mu

Fig. 4

Intra-sonographer repeatability (ICC_Repeat) and inter-sonographer reproducibility (ICC_Reprod) versus number of acquisitions: a AC, b LF Intercept (LFI), c LF Midband (LFM), d LF Slope, e Kappa, and f Mu

Fig. 5

Repeatability Bland-Altman plots based on mean of first 4 acquisitions for a AC (dB/cm-MHz), b LF Intercept (LFI in dB), c LF Midband (LFM in dB), d LF Slope (LFS in dB/MHz), e Kappa (unitless), and f Mu (unitless). Dashed lines show bias and lower and upper LoA (95%); numerical values for bias and lower and upper LoA are displayed below each plot

Fig. 6

Reproducibility Bland-Altman plots based on mean of first 4 acquisitions for a AC (dB/cm-MHz), b LF Intercept (LFI in dB), c LF Midband (LF in dB), d LF Slope (dB/MHz), e Kappa (unitless), and f Mu (unitless). Dashed lines show bias and lower and upper LoA (95%); numerical values for bias and lower and upper LoA are displayed below each plot