Keywords obstetrics - ultrasound - fetus
Introduction
The measurement of fetal crown-rump length (CRL) during early pregnancy ultrasound
examinations (11–14 weeks) is crucial for accurate fetal gestational age
determination [1 ]
[2 ], chromosomal abnormality risk assessment
[3 ]
[4 ]
[5 ], and fetal growth and
development evaluation [6 ]
[7 ]
[8 ].
However, inaccuracies in CRL measurements, which are influenced by fetal position,
can significantly impact clinical decision-making and potentially lead to adverse
pregnancy outcomes [9 ]. It is well-documented
that fetal hyperextension and hyperflexion can cause overestimation or
underestimation of CRL measurements, respectively [10 ]
[11 ].
The Fetal Medicine Foundation (FMF) defines the neutral fetal position as one where
the fetal head and spine form a straight line. However, a clear definition of the
CRL plane in relation to this position remains absent. The French College of Fetal
Echography (CFEF) [12 ], the International
Society of Ultrasound in Obstetrics and Gynecology (ISUOG) [13 ], and the INTERGROWTH-21st Project [14 ] advocate for CRL measurements when amniotic
fluid is observable between the fetus's chin and chest to avoid hyperflexion.
However, these guidelines fail to sufficiently address the hyperextension position.
Studies by Wanyonyi et al. [15 ] and Roux et al.
[16 ] have endeavored to assess fetal
position by examining the angles formed by fetal anatomical lines. While relatively
objective, these methods still demonstrate some inconsistency. Modern intelligent
technologies have advanced the precision and repeatability of fetal biometric
measurements [17 ]
[18 ]
[19 ].
However, in CRL assessments, these technologies often overlook the critical
influence of fetal posture [20 ]
[21 ]. Furthermore, the inherent subjectivity of
conventional assessment methods complicates their integration into intelligent
detection systems. Additionally, our study identified specific limitations in the
current assessment methodologies. For instance, although amniotic fluid between the
fetal chin and chest typically suggests a neutral position, anterior ([Fig. 1a ]) or posterior ([Fig. 1b ]) hip tilting may indicate a
hyperflexed or hyperextended position instead. On the other hand, a lack of amniotic
fluid in this area does not necessarily imply hyperflexion. The fetus might still
assume a neutral posture ([Fig. 1c ]).
Fig. 1 Examples that deviate from the fetal position classification
methods used in previous studies. (a ) Fetal hyperflexed position: The
fetus exhibits hyperflexion, but there is amniotic fluid present between the
fetal chin and chest, with the fetal hip flexed forward. (b ) Fetal
hyperextended position: The fetus is in a hyperextended position, but there
is amniotic fluid present between the fetal chin and chest, with the fetal
hip extended backward. (c ) Fetal neutral position: The fetus is in a
neutral position, with the head and spine nearly aligned in a straight line,
but there is no amniotic fluid between the fetal chin and chest.
Given these challenges, it is essential to develop an innovative approach that not
only meets the requirements for integration with intelligent technologies but also
enhances the accuracy of fetal posture assessment. At present, research on methods
suitable for integration into intelligent applications for the assessment of fetal
position in early pregnancy is limited. Our study seeks to introduce, validate, and
illustrate the potential of a novel quantitative method for fetal position
assessment, tailored for integration into intelligent applications, thereby
surmounting the limitations of current methods through improved accuracy,
objectivity, and repeatability. This approach is expected to advance the field by
offering a more reliable means of fetal assessment, ultimately contributing to
improved clinical outcomes.
Methods
Study design and data sources
This retrospective, single-center study was conducted using ultrasound images.
Between January 1 and August 31, 2022, 2,582 fetal CRL images from 2,520 early
pregnancy nuchal translucency (NT) examinations at a tertiary hospital formed
the pilot group. Additionally, 1,450 CRL images from 1,418 distinct cases
collected between September 1 and December 31, 2022 comprised the validation
group. Two experienced sonographers, each with over five years of prenatal
ultrasound examination experience, selected images meeting specific quality
criteria for this study. The inclusion criteria were: (1) Singleton pregnancies;
(2) Fetal CRL measurements between 45mm and 84mm; (3) Absence of noticeable
fetal structural anomalies and with normal NT measurements; (4) Fetal alignment
with the mid-sagittal plane; (5) Clear visibility of the fetal head, buttock,
and back skin. The exclusion criteria focused on low-resolution and unclear
images. These sonographers did not participate in the subsequent study. The
study received approval from the ethics committee of the hospital. Informed
consent from patients was unnecessary because of the retrospective nature of the
study. To ensure privacy, all images were anonymized by removing personal
identifiers before being utilized in the research.
Evaluation of fetal position
The evaluation of fetal position was conducted by three senior sonographers
(R1/R2/R3), each with over 15 years of fetal ultrasound examination experience.
They independently reviewed the selected images. Due to limitations in the
commonly used criteria for determining the fetal posture in the CRL plane, this
study implemented the FMF's method, typically used for the NT plane, which
we adapted for the CRL plane. According to this method, a neutral position is
identified when the fetal head and spine form a nearly straight line without
significant forward or backward curvature ([Fig.
1c ]). A forward curve of the fetal head and spine is categorized as
hyperflexion ([Fig. 1a ]), while a backward
curve is considered hyperextension ([Fig.
1b ]). In instances of divergent opinions among sonographers, a
collaborative discussion ensued until a consensus on the fetal position was
achieved. This consensus was then established as the reference standard for
fetal position in the studyʼs images.
Measurement of the cranial-dorsal-hip angle (∠CDH) of the fetus
Two ultrasound technicians (M1/M2), each with over three years of prenatal
ultrasound experience, independently conducted blinded measurements of the
cranial-dorsal-hip angle (∠CDH) of fetuses in the images. The measurements were
recorded for analysis. The specific technique used the fetal CRL measurement
line, a straight line from the top of the fetal head to the bottom of the hip,
to define points A (head) and B (hip). Additionally, a line perpendicular to the
CRL line was drawn from the midpoint of the fetal mandible, intersecting the
skin at the neck and back region, to establish point C. The ∠CDH was then
measured by connecting points A, C, and B ([Fig.
2 ]). To ensure reliability, a random 10% sample of the selected images
was re-measured by technician M1 using the same method after a two-week
interval. This process aimed to assess the consistency of the measurements.
Fig. 2 Measurement of the cranial-dorsal-hip angle (∠CDH). The
fetal CRL line allows for the determination of points A (head) and B
(hip). Point C determined by drawing a line perpendicular to the CRL
line from the midpoint of the fetal mandible, extending backward until
it intersects with the fetal neck and back skin. The ∠CDH is defined by
connecting points A, C, and B.
Classification of fetal position with ∠CDH method
The final value of the ∠CDH for this study was established by averaging the
measurements taken by both technicians. To define the reference range for ∠CDH
in fetuses in a neutral position, we analyzed the ∠CDH distribution within the
pilot group. Based on this analysis, the 95% reference range for a neutral
position was calculated. Fetal positions were then classified according to this
range: values falling below the minimum of this range were categorized as
hyperflexion (overbent), and those exceeding the maximum were considered
hyperextension (overextended). This classification criterion was applied to
categorize the fetal positions in both the pilot and validation groups based on
their ∠CDH values.
Assessment of fetal position with three established methods
To evaluate the accuracy of the ∠CDH method alongside other established methods
for determining fetal positions, two experienced ultrasound sonographers
(D1/D2), each with a decade of prenatal ultrasound practice, collaboratively
reviewed and documented fetal positions for both image groups. They employed the
techniques proposed by Ioannou et al., Wanyonyi et al., and Roux et al., herein
referred to as the Ioannou, Wanyonyi, and Roux methods, respectively.
The flowchart of the study is presented in [Fig.
3 ].
Fig. 3 Flowchart of the study. ∠CDH: cranial-dorsal-hip angle.
Statistical analysis
Statistical analyses were conducted using SPSS (version 26.0) software and
MedCalc (version 20.0) software. Descriptive statistics for metric data include
mean±SD for normally distributed data or median (IQR) for skewed data.
Categorical data are presented as counts and their respective percentages.
Statistical comparisons were made among the distribution of the general
characteristics using the independent Mann-Whitney U-test for continuous
variables and the χ2 test for categorical variables. Bland-Altman plots with 95%
limits of agreement and intra-class correlation coefficients (ICC) were used to
assess the agreement between measurements taken by M1 and M2, as well as within
M1. The 95% reference range of ∠CDH for fetal neutral position was generated
using the percentile method, taking the values from both sides. The weighted
Kappa value (k value) was employed to evaluate the concordance between
each assessment method and the reference standard. The κ value was categorized
as follows for interpretation: 0.01–0.20, indicating poor agreement; 0.21–0.40,
denoting fair agreement; 0.41–0.60, representing moderate agreement; 0.61–0.80,
signifying good agreement; and 0.81–1.0, reflecting very good agreement. A
significance level of p <0.05 was considered statistically
significant.
Results
Participant and general characteristics
In the pilot group, 2,186 fetal CRL images were analyzed, excluding 29 images
deviating significantly from the mid-sagittal plane, 351 with unclear boundaries
of the head, hip, and back, 11 with increased NT, and 5 damaged images.
Similarly, the validation group comprised 1,193 fetal CRL images, excluding 12
images with marked deviation from the mid-sagittal plane, 234 with indistinct
boundaries, 8 with increased NT, and 3 damaged images. No significant
differences were observed in the basic characteristics of the cases included in
both the pilot and validation groups (p>0.05). The distribution of maternal
ages, gestational weeks, CRL values, fetal positions, and ∠CDH values for both
groups is detailed in [Table 1 ].
Table 1 Case Characteristics Stratified by
Group.
Pilot group (n=2186)
Validation group (n=1193)
p -value
Characteristic
Maternal age (y)
30 (27.0–33.0)
30.0 (27.5–33.0)
0.460
Gestational weeks (w)
12+4 (12+2 –12+6 )
12+4 (12+1 –12+6 )
0.065
Fetal CRL (mm)
60.5 (56.0–64.9)
60.2 (56.0–64.3)
0.312
Fetal position
0.554
Hyperflexion
487 (22.3)
270 (22.6)
...
Neutral
1616 (73.9)
869 (72.8)
...
Hyperextension
83 (3.8)
54 (4.5)
...
∠CDH value
...
Measured by M1
124.5 (119.1–130.1)
124.5 (119.2–129.9)
0.897
Measured by M2
124.3 (119.0–129.9)
124.3 (119.1–129.7)
0.971
Data are given as n(%) and median (Q1-Q3); Abbreviations: CRL: crown-rump
length; ∠CDH: cranial-dorsal-hip angle.
Inter-observer and intra-observer agreement
[Fig. 4 ] presents Bland–Altman plots
illustrating the consistency analysis of ∠CDH measurements conducted between two
observers (M1/M2) and within the same observer (M1 at different times). In all
cases, the mean differences in measurement closely approximated zero, indicating
the absence of significant systematic measurement bias, both between different
observers and within the same observer. Specifically, the average difference in
∠CDH measurements between different observers was 0.14°, with 95% limits of
agreement (LOAs) ranging from±1.90° ([Fig.
4a ]). Conversely, when considering measurements within the same
observer, the average difference was merely 0.07°, and the 95% LOAs span±1.88°
([Fig. 4b ]).
Fig. 4 Bland-Altman plots with 95% limits of agreement (LoA) of
inter-observer agreement (a ) and intra-observer (b ) of
∠CDH measurements. The orange dashed line represents the zero reference
line, the blue solid line represents the mean difference, the green
vertical lines represent the 95% confidence interval of the mean
difference, the dark brown dashed line represents the LoA line, the blue
vertical lines represent the 95% confidence interval of LoA, and the
purple dashed line represents the regression line for the pair
difference.
The absolute agreement in measurements was notably high, as evidenced by the
intraclass correlation coefficient (ICC) values of 0.993 ( 95%CI:0.992, 0.993;
p<0.001) for inter-observer measurements and 0.993 ( 95%CI:0.992, 0.995;
p<0.001) for intra-observer measurements.
Accuracy of ∠CDH method for determining fetal position
In the pilot group, the distribution of ∠CDH values for fetuses identified in the
neutral position ranged from 112.6° to 142.6°, with a median of 126.5° and an
interquartile range of 122.6° to 130.8°. The 95% reference range for neutral
position ∠CDH values was established at 118.3° to 137.8°. Using this range, 489
images (22.4%) were classified as hyperflexion, 1578 images (72.2%) as neutral,
and 119 images (5.4%) as hyperextension. The accuracy of this classification
method in the pilot group was 94.5%, demonstrating high agreement with the
reference standard (k value=0.874; 95% CI: 0.852, 0.896; P<0.001).
In the validation group, based on the same criteria, 266 images (22.3%) were
classified as hyperflexion, 845 images (70.8%) as neutral, and 82 images (6.9%)
as hyperextension. The classification accuracy in the validation group was
92.6%, with a high level of agreement with the reference standard (k
value=0.838; 95% CI: 0.806, 0.871; P<0.001).
Comparison between the ∠CDH method and other fetal position determination
methods
In the pilot group, the ∠CDH method demonstrated superior accuracy at 94.5%,
surpassing the Ioannou, Wanyonyi, and Roux methods, which showed accuracies of
82.3%, 82.7%, and 85.0% respectively. The corresponding k values were 0.874 for
the ∠CDH method, and 0.461, 0.560, and 0.575 for the other methods, all with
P-values<0.001. The validation group showed a similar trend, with the ∠CDH
method achieving an overall accuracy of 92.6%, higher than the 80.5%, 81.1%, and
83.7% accuracies of the Ioannou, Wanyonyi, and Roux methods, respectively. The
kappa values were 0.838 for the ∠CDH method and 0.408, 0.534, and 0.550 for the
others (all P<0.05). [Fig. 5 ] presents
some examples of fetal position assessments using these four methods. Notably,
the ∠CDH method's accuracy in identifying hyperflexion positions stood out,
with rates of 92.4% in the pilot group and 88.1% in the validation group,
significantly higher than the 46.2% and 40.4% accuracy rates of the other three
methods, as shown in [Table 2 ]
[3 ].
Fig. 5 Examples of four methods for assessing fetal positions.
(a ) Neutral position. There is amniotic fluid between the
fetal chin and chest, and both the profile line and palate form acute
angles with the CRL line, ∠CDH=129.0°. (b ) Neutral position.
There is no amniotic fluid between the fetal chin and chest, and both
the profile line and palate form acute angles with the CRL line,
∠CDH=129.6°. (c ) Neutral position. There is amniotic fluid
between the fetal chin and chest, the profile line does not intersect
with the CRL line in front of the fetal buttock, and the palate
intersects with the CRL line at an angle<90°, ∠CDH=134.9°. (d )
Hyperflexed position. There is no amniotic fluid between the fetal chin
and chest, and both the profile line and palate form acute angles with
the CRL line, ∠CDH=110.8°. (e, f ) Hyperflexed position. There is
amniotic fluid between the fetal chin and chest, and both the profile
line and palate form acute angles with the CRL line, ∠CDH=109.7°,
110.8°; (g ) Hyperextended position. There is amniotic fluid
between the fetal chin and chest, the profile line is nearly parallel to
the CRL line, and the palate forms a 90° angle with the CRL line,
∠CDH=151.3°. (h, i ) Hyperextended position. There is amniotic
fluid between the fetal chin and chest, the profile line is nearly
parallel to the CRL line, and the palate forms an angle<90° with the
CRL line, ∠CDH=152.1°, 149.2°. The yellow dashed line represents the CRL
line, the blue dashed line represents the profile line.
Table 2 Comparison of the ∠CDH method with other methods
for fetal position classification in the pilot group.
Reference standards
Method Ioannou
Method Wanyonyi
Method Roux
Method ∠CDH
Neutral (n=1616)
Neutral
1574 (97.5)
1501 (92.9)
1572 (97.3)
1537 (95.1)
Hyperflexion
41 (2.5)
41 (2.5)
41 (2.5)
39 (2.4)
Hyperextension
0 (0.0)
74 (4.6)
3 (0.2)
40 (2.5)
Hyperflexion (n=487)
Hyperflexion
225 (46.2)
225 (46.2)
225 (46.2)
450 (92.4)
Neutral
262 (53.8)
262 (53.8)
262 (53.8)
37 (7.6)
Hyperextension
0 (0.0)
0 (0.0)
0 (0.0)
0 (0.0)
Hyperextension (n=83)
Hyperextension
0 (0.0)
82 (98.8)
60 (72.3)
79 (95.2)
Neutral
82 (98.8)
0 (0.0)
22 (26.5)
4 (4.8)
Hyperflexion
1 (1.2)
1 (1.2)
1 (1.2)
0 (0.0)
Total (n=2186)
Correct
1800 (82.3)
1808 (82.7)
1857 (85.0)
2066 (94.5)
Incorrect
386 (17.7)
378 (17.3)
329 (15.0)
120 (5.5)
k-value
0.461 (0.419–0.503)
0.560 (0.522–0.599)
0.575 (0.534–0.616)
0.874 (0.852–0.896)
p-value
--a
<0.001b
<0.001c
Data are given as n(%); k -values are given as value (95%CI); in
the p -values, a represents the comparison between the
Ioannou method and the ∠CDH method; b represents the
comparison between the Wanyonyi method and the ∠CDH method; c
represents the comparison between the Roux method and the ∠CDH
method.
Table 3 Comparison of ∠CDH method with other methods for
fetal position classification in the validation group.
Reference standards
Method Ioannou
Method Wanyonyi
Method Roux
Method ∠CDH
Neutral (n=869)
Neutral
851 (97.9)
804 (92.5)
850 (97.8)
813 (93.6)
Hyperflexion
18 (2.1)
18 (2.1)
18 (2.1)
28 (3.2)
Hyperextension
0 (0.0)
47 (5.4)
1 (0.1)
28 (3.2)
Hyperflexion (n=270)
Hyperflexion
109 (40.4)
109 (40.4)
109 (40.4)
238 (88.1)
Neutral
161 (59.6)
161 (59.6)
161 (59.6)
32 (11.9)
Hyperextension
0 (0.0)
0 (0.0)
0 (0.0)
0 (0.0)
Hyperextension (n=54)
Hyperextension
0 (0.0)
54 (100.0)
40 (74.1)
54 (100.0)
Neutral
54 (100.0)
0 (0.0)
14 (25.9)
0 (0.0)
Hyperflexion
0 (0.0)
0 (0.0)
0 (0.0)
0 (0.0)
Total (n=1193)
Correct
960 (80.5)
967 (81.1)
999 (83.7)
1105 (92.6)
Incorrect
233 (19.5)
226 (18.9)
194 (16.3)
88 (7.4)
k-value
0.408 (0.351–0.465)
0.534 (0.482–0.587)
0.550 (0.494–0.605)
0.838 (0.806–0.871)
p-value
--a
<0.001b
<0.001c
Data are given as n(%); k -values are given as value (95%CI); in
the p -values, a represents the comparison between the
Ioannou method and the ∠CDH method; b represents the
comparison between the Wanyonyi method and the ∠CDH method; c
represents the comparison between the Roux method and the ∠CDH
method.
Discussion
The principal finding of the study is the establishment of the ∠CDH as a reliable
and
accurate method for fetal position assessment in early pregnancy ultrasound
examinations. This method was demonstrated to be highly consistent, as evidenced by
ICCs of 0.993. The ∠CDH method proved to be more accurate than existing methods,
with the neutral fetal positionʼs ∠CDH range identified as 118.3° to 137.8° and
exhibited an approximate 10% increase in accuracy in both the pilot and validation
groups.
A key strength of the ∠CDH method is its pioneering role in introducing a
quantitative approach to fetal position assessment. Although this study did not
engage in experiments using this angle for intelligent fetal position assessment,
it
successfully validated the method's reliability and accuracy for the first
time, indisputably laying a solid foundation for future integration with intelligent
technologies. Additionally, the ∠CDH method offers a more comprehensive assessment
by accounting for the fetal hip's position relative to the head and spine,
simplifying and directly correlating the measured angle with the corresponding
posture. This innovative approach not only fills a critical gap in existing
methodologies but also enhances the method's objectivity.
Comparison with existing literature reveals the innovative nature of the ∠CDH method
in evaluating fetal posture. The Ioannou method [14 ] is renowned for its high accuracy in identifying neutral positions,
as evidenced by its impressive performance in our study: 97.5% accuracy in the pilot
group and 97.9% in the validation group. However, it falls short in recognizing
hyperextension, often incorrectly classifying these cases as neutral. The methods
of
Wanyonyi et al. [15 ] and Roux et al. [16 ], although innovative in their angular
approach, fail to independently confirm the fetal neutral and hyperflexed position
(in which both of these angles measure less than 90°) and remain contingent upon the
detection of amniotic fluid between the fetal chin and chest. This critical
dependency may result in misclassification when faced with scenarios involving the
nuanced role of hip orientation, as previously discussed. The ∠CDH method recognizes
that the hip's orientation relative to the head and spine can dramatically
influence fetal position and, consequently, CRL measurement. By integrating the
position of the hip, this method transcends the limitations of the aforementioned
methods, providing a more robust and comprehensive framework for fetal position
classification. This is particularly critical in cases where traditional methods are
prone to misclassification, thereby reducing the margin of error and increasing the
reliability of fetal assessments in early pregnancy.
In clinical practice, it is crucial to acknowledge that images of fetuses in
hyperextended positions are significantly less common compared to those in neutral
or hyperflexed positions. This rarity often stems from the fact that a hyperextended
fetal body frequently does not align with the mid-sagittal plane, which clinicians
often prioritize over fetal posture. Consequently, if a fetus is not positioned in
the mid-sagittal plane, sonographers tend not to retain the image. As a result, the
classification of neutral and hyperflexed positions becomes more significant and
practical in everyday clinical settings. Our findings reveal that the ∠CDH method
exhibits superior accuracy in differentiating these two postures. In the pilot
group, it achieved 95.1% accuracy for the neutral position and 92.4% for the
hyperflexed stance, while in the validation group, accuracy rates were 93.6% for the
neutral posture and 88.1% for the hyperflexed position. This precision in commonly
encountered scenarios underscores the practical value of our method in clinical
applications.
Our approach also acknowledges the dynamic nature of fetal movement, where positions
can fluctuate during the scanning process. This consideration, which is often
underemphasized in the literature, has significant clinical relevance. Utilizing the
∠CDH as a quantitative measure enables a reduction in the reliance on subjective
interpretations of fetal posture. Such interpretations, which can vary depending on
the sonographer's experience, may result in inconsistencies, as indicated in
previous studies on similar assessment tasks [22 ]
[23 ].
Insights from the field highlight a critical demand for standardization and
objectivity in fetal measurements [24 ]. The
∠CDH method, with its quantification capabilities, not only satisfies this
requirement but also harmonizes with the evolving integration of intelligent
detection systems in prenatal diagnostics. The adoption of the ∠CDH method in
automated measurement systems has the potential to foster significant advancements
in the field, as observed by Cengiz, et al. in their study on automated CRL
measurement techniques [21 ].
Our study has certain limitations including its single-center design and a limited
sample size for hyperextended positions. This aspect could potentially affect the
generalizability of our findings. Moreover, the study did not directly explore the
impact of varying fetal positions on CRL measurements, highlighting an area in need
of further investigation.
Conclusion
To summarize, the introduction of the ∠CDH method is a pivotal development in fetal
position assessment during early pregnancy. Its ability to provide a quantitative,
objective tool for this purpose holds great clinical relevance. The potential for
this method's integration into automated technologies bodes well for the future
of obstetric care, promising to enhance the precision of CRL measurements and
improve overall clinical outcomes. Future research, particularly multi-center
studies that encompass a wider variety of fetal positions, will be crucial in
further validating and refining the ∠CDH method.
Bibliographical Record
