Comparative Study of 2D vs. 3D AI-Enhanced Ultrasound for Accurate Fetal Crown–rump Length Evaluation in the First Trimester

Comparative Study of 2D vs. 3D AI-Enhanced Ultrasound for Accurate Fetal Crown–rump Length Evaluation in the First Trimester | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Comparative Study of 2D vs. 3D AI-Enhanced Ultrasound for Accurate Fetal Crown–rump Length Evaluation in the First Trimester Yuanji Zhang, Yuhao Huang, Chaoyu Chen, Xing Hu, Wenxiong Pan, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6250944/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 16 Jul, 2025 Read the published version in BMC Pregnancy and Childbirth → Version 1 posted 10 You are reading this latest preprint version Abstract Background Accurate fetal growth evaluation is crucial for monitoring fetal health, with crown-rump length (CRL) being the gold standard for estimating gestational age and assessing growth during the first trimester. To enhance CRL evaluation accuracy and efficiency, we developed an AI-based model (3DCRL-Net) using the 3D U-Net architecture for automatic landmark detection to achieve CRL plane localization and measurement in 3D ultrasound. We then compared its performance to that of experienced radiologists using both 2D and 3D ultrasound for fetal growth assessment. Materials and methods This prospective cohort study collected fetal data from 1,326 ultrasound screenings conducted between 11 and 14 weeks of gestation (June 2021 to June 2023). Three experienced radiologists performed fetal screening using 2D video (2D-RAD) and volumes (3D-RAD) to obtain the CRL plane and measurement. The 3DCRL-Net model automatically outputs the landmark position, CRL plane localization and measurement. Three specialists audited the planes achieved by radiologists and 3DCRL-Net as standard or non-standard. The performance of CRL landmark detection, plane localization, and measurement was evaluated in the internal testing dataset, comparing results with 3D-RAD. The performance of CRL plane localization, measurement, and time efficiency were assessed between 3DCRL-Net, 2D-RAD, and 3D-RAD in the external dataset. Results The internal dataset consisted of 126 cases in the testing set (training: validation: testing = 8:1:1), and the external dataset included 245 cases. On the internal testing dataset, 3DCRL-Net demonstrated a mean absolute distance error of 1.81 mm for the nine landmarks, high accuracy in standard plane localization compared to 3D-RAD (91.27% vs. 80.16%), and strong consistency in CRL measurements, with a 1.26 mm measurement error ( P = 0.70). On the external testing dataset, 3DCRL-Net demonstrated high performance in standard plane localization, achieving results comparable to 2D-RAD and 3D-RAD (accuracy: 91.43% vs. 91.43% vs. 86.12%), with strong consistency in CRL measurements, compared to 2D-RAD, which showed a mean absolute error of 1.58 mm and a mean difference of 1.12 mm ( P = 0.25). Conclusions The 3DCRL-Net model achieved high performance in CRL evaluation of the standard plane, providing reliable CRL measurements comparable to radiologists using 2D ultrasound and surpassing those using 3D ultrasound, while facilitating CRL evaluation in clinical practice within a few seconds. Crown-Rump Length CRL Evaluation Artificial Intelligence Fetal Growth Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1 Background Accurate fetal growth evaluation is crucial for monitoring fetal health and predicting the risk of anomalies [ 1 ]. The crown-rump length (CRL) is considered the gold standard for estimating gestational age (GA) and assessing fetal growth during the first trimester [ 2 , 3 ]. A precise CRL measurement aids in diagnosing fetal growth restriction, enables accurate evaluation of nuchal translucency (NT) for assessing the risk of aneuploidies, and enhances the detection of structural anomalies [ 4 – 9 ]. The standard CRL plane and measurements must meet strict quality control standards to ensure diagnostic accuracy, but CRL evaluation depends heavily on the radiologist's experience [ 10 – 15 ]. Identifying the optimal fetal position and acquiring the correct plane during two-dimensional (2D) screening is time-consuming and prone to errors [ 16 , 17 ]. Significant variability between radiologists can further lead to diagnostic inaccuracies [ 18 , 19 ]. While three-dimensional (3D) ultrasound allows offline manipulation for accurate CRL measurement, it remains challenging due to the complexity of data processing and the need for familiarity with the system [ 20 , 21 ]. Artificial intelligence (AI) has demonstrated potential in predicting GA from 3D data in the first trimester [ 22 – 24 ]. However, limited AI validation in CRL analysis, including plane localization and measurement accuracy, comparison with traditional methods (e.g., 2D screening), and improved interpretability, hampers clinician trust and adoption of DL-based systems [ 25 – 27 ]. In this study, we developed an AI-based model using the 3D U-Net (named 3DCRL-Net). It accurately detects multiple landmarks to regress the standard plane and measure CRL for assessing fetal growth during the first trimester. We also compared its performance with experienced radiologists using both 2D and 3D ultrasound screening to explore its potential in clinical practice. 2 Materials and Methods 2.1 Study design and data collection In this prospective cohort study, fetal data, including 2D videos and 3D volumes, were collected from 1326 fetal ultrasound screenings conducted between 11 and 14 weeks of gestation at the Third Affiliated Hospital of Shenzhen University, from June 2021 to June 2023. The study was approved by the Medical Research Ethics Committee (Approval No. 2019-LHQRMYY-LL-062). Written informed consent was obtained from all participating pregnant women. The inclusion criteria for the data were singleton fetuses with a CRL between 45 mm and 84 mm, without structural abnormalities, and a fetal position conducive to obtaining CRL plane. All ultrasound examinations were performed using the following equipment: General Electric Voluson E6/E8/E10 (GE Healthcare, Zipf, Austria), Philips EPIQ7 (Philips, Seattle, Washington, USA), and Mindray Resona-7 (Mindray, Shenzhen, Guangdong Province, China), with a 3D transabdominal volume probe. The standard CRL plane should meet the five criteria outlined by the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) [ 8 ]: midsagittal plane, neutral position, horizontal orientation (crown-rump line at 75° to 105°), clear visibility of the crown and rump, and adequate magnification (over two-thirds of the image). For each fetus, the radiologists adjusted the volume box to encompass the crown and rump for 3D acquisition and ensured proper adjustment of 2D ultrasound videos to obtain the standard CRL images. Cases without optimal CRL within 15 minutes were excluded, and successful images and volumes were saved for analysis. The time cost for both 2D video and 3D volume was recorded. Quality control was performed by three radiology specialists with 20–30 years of experience in fetal examination to ensure the presence of the standard CRL plane in all fetal data. Only data confirmed by at least two radiologists were included in the study. The data collection process is shown in Fig. 1 . 2.2 Annotation of internal and external dataset In the internal dataset, a total of 1,000 fetal volumes were collected and divided into training, validation, and testing sets in an 8:1:1 ratio. For CRL analysis and model development, each case was annotated with nine landmarks, a standard CRL plane, and a CRL measurement. The nine structural landmarks included the cranial crest, thalamus, diencephalon, nasal bone, lower alveolar, chest wall, diaphragm, lumbar, and buttocks. Besides, the standard CRL plane equation ( \(\:ax+by+cz+d=0\) ) and CRL measurement were also annotated for model evaluation. This process was facilitated using the Pair Annotation Software Package 3.0 [ 28 ] by three experienced radiologists (> 5 years of fetal examination experience). Three radiology specialists with 20–30 years of experience performed quality control on the annotations made by the experienced radiologists. In the external dataset, 245 fetal cases, including both 2D and 3D data, were collected. To compare the performance of 3DCRL-Net vs. radiologists, the same three radiologists independently localized the standard CRL plane in both the 2D and 3D data, measured the CRL, and recorded the time taken to identify the plane and measure the CRL for each case. 2.3 The framework of 3DCRL-Net Model The framework of 3DCRL-Net model was shown in Fig. 2 . Building on our previous work in 3D fetal ultrasound anatomical landmark detection (FetusMapV2 [ 29 ]), we developed a simplified version for CRL analysis. The proposed framework mainly includes a c landmark detector, taking the 3D fetus as input and outputting \(\:k\) CRL-related landmarks with Gaussian heatmap representation. During training, we leveraged the structure-unconstrained gradient checkpoint (GCP) and activation-unreserve reversible layer (REV) techniques to reduce the memory requirements and enable the low-resource training on limited GPU memory. This supports high-resolution input, finally enhancing the landmark detection accuracy. The Kullback-Leibler (KL) divergency loss was used to optimize the learning process by calculating the distance between annotated and predicted heatmap distributions. During testing, the detector takes the entire 3D volume as input and generates Gaussian heatmaps of landmarks as output. Then, the position with the highest value of each heatmap will be considered as the final landmark prediction (i.e., [x, y, z]). Based on the detected landmarks: (a) We used the Ordinary Least Squares (OLS) method to regress the plane equation. The goal is to minimize the sum of squared Euclidean distances from these landmarks to the plane. First, we calculate the center of these landmarks ( \(\:\stackrel{-}{x},\stackrel{-}{y},\stackrel{-}{z}\) ) and decentralize them to eliminate positional influence, \(\:{r}_{i}=({x}_{i}-\stackrel{-}{x},{y}_{i}-\stackrel{-}{y},{z}_{i}-\stackrel{-}{z})\) . Next, we construct the covariance matrix \(\:(U=\sum\:{r}_{j}{r}_{j}^{T})\) and perform eigenvalue decomposition ( \(\:Uv=\lambda\:v\) ). We then select the eigenvector \(\:{v}_{min}=(a,b,c)\) corresponding to the smallest eigenvalue as the plane's normal vector. Given that the plane passes through the center point, the offset \(\:d\) of the plane equation is calculated as: \(\:d=-(a\stackrel{-}{x}+b\stackrel{-}{y}+c\stackrel{-}{z}))\) . Finally, the CRL plane equation in 3D space can be defined as: $$\:ax+by+cz+d=0.$$ (b) We measure the CRL by calculating the distance between two landmarks (cranial crest and buttocks). 2.4 Experiments on Internal and External Testing Datasets We implemented our proposed landmark detector in Python and Pytorch, using one NVIDIA GTX 2080Ti GPU with 12 GB memory. All volumes were resized to 208*128*144 to maximize the memory usage. We used Adam optimizer with learning rate of 1e-4 and moment term of 0.5, to update the model weights. The total training epoch is 100 and batch size is 1. For the standard CRL plane evaluation, three radiology specialists with 20–30 years of experience audited the accuracy of CRL plane localization for the 3DCRL-Net, 2D-radiologist (2D-RAD), and 3D-radiologist (3D-RAD) groups. Each image was assessed and assigned a binary score: 0 (non-standard plane) and 1 (standard plane). Only results where at least two specialists agreed on the same score (either 0 or 1) were considered valid. 1) The Performance of 3DCRL-Net in Internal Testing Datasets The performance of the 3DCRL-Net model in the internal testing dataset was evaluated for CRL landmarks, plane localization, and measurement. The 3DCRL-Net model's results were compared with 3D-RAD for the nine key landmarks in distance error assessment, CRL plane localization accuracy using both subjective evaluations by specialists and objective comparisons, and CRL measurement results on the standard plane. Time efficiency was also compared between the two groups. 2) Accuracy of 3DCRL-Net in External Testing Datasets The external dataset was used to validate the accuracy and efficiency of the 3DCRL-Net model for CRL plane localization and measurements. The performance of 3DCRL-Net in automatically identifying and measuring CRL was compared with 2D-RAD and 3D-RAD, audited by three specialists. Time efficiency was also compared across the radiologist groups. 2.5 Statistical analysis To assess the performance of CRL structural landmarks, plane localization and measurement, several metrics were employed, including mean, mean absolute error, mean difference, standard deviation (SD), and confidence intervals (CI). Additionally, plane angular deviation and landmark distance errors were calculated. Interobserver variability was assessed using Bland-Altman plots. Differences in measurement techniques were analyzed using chi-square tests and T-tests. Time costs were also evaluated. All statistical analyses were conducted using Python 3.11 (Python Software Foundation), with a significance threshold of P < 0.05. 3 Results 3.1 Characteristics of the study The internal training and validation dataset consisted of 748 and 126 fetal volumes, respectively, the internal testing dataset contained 126 fetal volumes, and the external dataset comprised 245 fetal videos and volumes. The average gestational age of the fetuses in the internal testing dataset and external dataset was 12 ± 3 weeks, and 12 ± 2 weeks, respectively. 3.2 Performance of 3DCRL-Net model on internal testing dataset 3.2.1 Assessment of structural landmarks The performance of the 3DCRL-Net model in predicting landmarks was evaluated by comparing its results with 3D-RAD, as shown in Table 1 . The average absolute distance error for the nine landmarks was 1.81mm, with a SD of 1.10mm. Specifically, the error for the cranial-crest was 1.68 ± 0.82 mm, and for the buttocks, it was 2.02 ± 2.06mm. These results indicate that the 3DCRL-Net model can accurately identify the cranial-crest and buttocks landmarks, laying a solid foundation for measuring CRL in 3D volume. Table 1 Summary of Internal Testing Dataset Experimental Results for Nine Landmarks Localization Structural Landmark Distance Error (mean ± SD) (mm) Cranial crest 1.68 ± 0.82 Thalamus 1.54 ± 0.69 Diencephalon 1.63 ± 0.82 Nasal bone 1.54 ± 0.99 Lower alveolar 1.50 ± 0.71 Chest wall 2.02 ± 1.11 Diaphragm lumbar 1.72 ± 0.92 Umbilical 2.62 ± 1.79 Buttocks 2.02 ± 2.06 Mean 1.81 ± 1.10 3.2.2 Objective and subjective assessment of CRL plane To assess the accuracy of the predicted CRL planes derived from landmark regression, we conducted both objective and subjective evaluations on 126 cases, as shown in Table 2 . The objective evaluation measured the spatial differences between the 3DCRL-Net and 3D-RAD. Specifically, we computed the angular deviation with a precision of 5.34° to quantify the discrepancy between the two planes. For the subjective evaluation audited by specialists, the 3DCRL-Net model achieved an accuracy of 91.27% (115 standard CRL planes), while the 3D-RAD had an accuracy of 80.16% (101 cases met the criteria). 3.2.3 Assessment of CRL measurement The performance of 3DCRL-Net in measuring CRL on the standard plane is presented in Table 2 . The CRL measurement in 3D is defined as the distance between the cranial-crest and the rump landmarks. The results show that the mean and SD of CRL measurements obtained by 3DCRL-Net and 3D-RAD are 63.06mm ± 8.13mm and 62.60mm ± 7.93mm, respectively, with no significant difference between them ( P = 0.70). Moreover, due to the 3DCRL-Net 's accurate prediction of the cranial-crest and rump landmarks, the CRL measurement estimated by the 3DCRL-Net closely matches the 3D-RAD results, with a mean absolute error of 1.26mm ± 1.14mm. We recorded the total time from inputting the 3D volume to outputting the CRL measurement for both 3DCRL-Net and 3D-RAD. The mean time cost for fetal cases was 2.02 seconds for 3DCRL-Net and 10 minutes for 3D-RAD ( P < 0.001). Table 2 Objective and Subjective Evaluations of CRL Plane Localization and Measurement Results CRL plane Angle Error (°) 3DCRL-Net Accuracy (%) 3D-RAD Accuracy (%) 5.34 91.27 80.16 CRL measurement Measurement Mean Absolute Error (mean ± SD) (mm) 3DCRL-Net Measurement (mm) 3D-RAD Measurement (mm) 1.26 ± 1.14 63.06 ± 8.13 62.60 ± 7.93 3.3 Performance of 3DCRL-Net model on external testing dataset 3.3.1 Comparative Performance in CRL Plane Localization: 3DCRL-Net vs. 2D-RAD vs. 3D-RAD For CRL plane localization, the comparative results from the specialists' audit of the 3DCRL-Net, 2D-RAD, and 3D-RAD groups, including both standard and non-standard planes, are summarized in Table 3 . The 3DCRL-Net model achieved an accuracy of 91.43%, identifying 224 standard CRL planes out of 245 fetal volumes, while the 2D-RAD group identified 228 standard planes (93.06% accuracy), and the 3D-RAD group identified 211 standard planes (86.12% accuracy). The most common reason for a non-standard CRL plane in the 3DCRL-Net group was "Clear Visibility of Crown and Rump" whereas in the 2D-RAD of "Midsagittal Plane" and 3D-RAD groups, it was "Clear Visibility of Crown and Rump". The examples of standard and non-standard CRL planes obtained by the three groups are shown in Fig. 3 . Table 3 Accuracy of CRL Plane Localizations by 3DCRL-Net and Radiologists on the External Testing Dataset. The accuracy of ultrasound (US) images was assessed by radiology specialists. A standard image met all five criteria, while a non-standard image was assigned specific reasons. Image Result No. of Cases 3DCRL-Net 2D-RAD 3D-RAD Standard 224 (91.43%) 228 (93.06%) 211(86.12%) Non-standard 21 (8.57%) 17 (6.94%) 34 (13.88%) Reason Midsagittal Plane 6 7 11 Neutral Position 3 4 1 Horizontal Orientation 4 3 2 Clear Visibility of Crown and Rump 7 2 19 Adequate Magnification 1 1 1 3.3.2 Comparative Performance in CRL measurement: 3DCRL-Net vs. 2D-RAD vs. 3D-RAD For CRL measurement, the mean and SD of the standard CRL measurements for the 3DCRL-Net, 2D-RAD, and 3D-RAD groups were 59.90 mm ± 6.93mm, 61.05 mm ± 7.51mm, and 60.02 mm ± 7.40mm, respectively. The mean absolute error of CRL measurement between 2D-RAD and 3DCRL-Net was 1.58 mm ± 1.43 mm, and between 2D-RAD and 3D-RAD was 1.37 mm ± 0.95 mm, with no significant differences observed between 2D-RAD and 3DCRL-Net ( P = 0.25), or between 2D-RAD and 3D-RAD ( P = 0.19, ANOVA). The agreement distribution of CRL measurements between the three groups is shown in the Bland-Altman plot (Fig. 4 ). The mean differences and SD between 2D-RAD and 3DCRL-Net, 2D-RAD and 3D-RAD are 1.12 mm ± 3.57mm, and 1.06 mm ± 2.66mm, respectively. Moreover, the mean, median, and 95% CI results across different weeks (11–14 weeks) are shown in Table 4 and Fig. 5 . We recorded the total time from scanning the 3D volume to outputting CRL measurements for 3DCRL-Net and 3D-RAD, and from scanning 2D screening to measuring CRL for 2D-RAD. The mean time cost was 5 seconds for 3DCRL-Net (1.02 seconds for model prediction), 11 minutes for 3D-RAD, and 35 seconds for 2D-RAD (P < 0.001). Table 4 The Mean, Median, and 95% CI of Standard CRL Measurements for the 3DCRL-Net, 2D-RAD, and 3D-RAD Groups Across Weeks 11–14. Since there was only one case at 14 + 0 weeks, it was included in the 13-week group for statistical purposes. Groups 3DCRL-Net 2D-RAD 3D-RAD 11 Weeks Mean(mm) 49.70 50.23 49.21 Median(mm) 49.67 50.14 49.06 95%CI (mm) [48.98, 50.42] [49.57, 50.90] [48.40, 50.02] 12 Weeks Mean(mm) 58.96 59.85 58.92 Median(mm) 59.24 60.10 59.45 95%CI (mm) [58.38, 59.54] [59.29, 60.41] [58.33, 59.51] 13 Weeks Mean(mm) 70.46 72.95 71.56 Median(mm) 70.57 71.54 71.35 95%CI (mm) [69.14, 71.79] [71.45, 74.45] [70.22, 72.90] 4 Discussion In this study, we developed an AI-based model, named 3DCRL-Net, to acquire the standard CRL plane and accurately measure CRL using 3D volume data during 11 to 14 weeks. The 3DCRL-Net demonstrated high accuracy in identifying structural landmarks (mean absolute distance error, 1.81mm), locating the standard CRL plane (accuracy, 91.27%), and measuring CRL (mean absolute error, 1.26mm) in the internal testing dataset, audited by specialists. In a comparative experiment with experienced radiologists using 2D and 3D ultrasound on the external dataset, 3DCRL-Net (91.43%) achieved comparable accuracy in CRL plane localization to 2D-RAD (93.06%), slightly higher than 3D-RAD (86.12%). The model also showed minimal differences in CRL measurement compared to radiologists using 2D ultrasound, with a mean absolute error of 1.58mm and mean difference of 1.12 mm. These results suggest that 3DCRL-Net has strong potential for fetal crown-rump length evaluation in the first trimester. Accurate CRL measurement is essential for fetal growth assessment and clinical management [ 1 , 2 , 10 ]. In practice, it relies on radiologists' ability to precisely locate the standard plane, demanding extensive knowledge and skill [ 3 ]. Previous studies have highlighted the global challenge of significant variability among radiologists, leading to inaccurate CRL plane acquisition and measurement [ 7 , 10 – 13 ]. Traditional CRL plane localization and measurement in 2D screening are prone to errors and are time-consuming for obtaining the standard plane. In 3D volumes, expertise is still required to handle high-dimensional data and spatial awareness, presenting an ongoing challenge [ 18 , 21 , 30 ]. Therefore, this study developed the 3DCRL-Net to automatically detect anatomical structures in the 3D fetal volume, minimizing the technical expertise required from the radiologist. It located the standard CRL plane and performed measurements based on the detected landmarks. This model provided reliable and efficient results, reducing operator-related variability and improving the accuracy of fetal growth evaluation. Compared to 2D ultrasound, the enhanced volumetric data from 3D ultrasound significantly improves the detection of key anatomical landmarks, aiding in the localization of the standard CRL plane and enhancing the accuracy of CRL measurements for fetal growth evaluation [ 30 ]. In our 3DCRL-Net, we determine the standard CRL plane by detecting nine key structural landmarks on the midsagittal plane, rather than relying solely on single-plane localization within the 3D volume. This approach helps mitigate biases and measurement variability caused by subjective radiologist interpretation of plane identification. In the task of detecting the nine landmarks, the 3DCRL-Net model demonstrated high consistency with experienced radiologists using 3D ultrasound, indicating its strong performance in landmark detection. For measuring CRL at two key anatomical points, the mean absolute distance for the cranial crest and buttocks in 3D volume was less than 2 mm. For CRL plane localization assessment, our 3DCRL-Net model demonstrated high accuracy, comparable to experienced radiologists using both 2D and 3D ultrasound in standard CRL plane localization. In the external dataset, 3DCRL-Net achieved performance comparable to 2D-RAD and outperformed 3D-RAD, with a 5.31% improvement in accuracy. This highlights the 3DCRL-Net model's effectiveness in locating the standard CRL plane in 3D volumes through key structural landmark detection, making it suitable for clinical practice. Radiologists were more likely to fail in identifying the midsagittal plane with 2D ultrasound, and both the midsagittal plane and crown-rump visibility with 3D ultrasound. In contrast, the 3DCRL-Net model mainly struggled with crown-rump visibility. These challenges highlight the difficulty in accurately identifying the midsagittal plane. Fetal posture issues during 3D volume scanning, such as crown-rump visibility, neutral position, and horizontal orientation, can lead to non-standard CRL plane localization. For the comparative results of CRL measurement in the standard CRL plane, the 3DCRL-Net demonstrated high consistency with radiologists using 2D ultrasound, with mean differences of 1.12mm. Furthermore, in comparison box plots of 3DCRL-Net, 2D-RAD, and 3D-RAD at 11, 12, and 13 weeks, both the mean and median CRL measurements showed high consistency, demonstrating the potential of 3DCRL-Net to accurately identify key anatomical landmarks and measure CRL across different gestational weeks. The time results further highlighted the great potential of 3DCRL-Net in clinical settings, providing precise CRL measurements while saving time in fetal screening—saving 0.5 minutes compared to 2D-RAD and 10 minutes compared to 3D-RAD. This could potentially transform clinical scanning practices by alleviating the high technical demands on radiologists. Although the 3DCRL-Net model automatically locates the CRL plane and measures CRL with high performance, there are still limitations to consider. First, we did not evaluate the impact of different clinicians' techniques on 3D volume collection, which may require proper training of radiologists to ensure correct fetal positioning. Additionally, we did not assess the model's performance in difficult clinical cases, such as maternal conditions or fetal abnormalities. These aspects will be addressed in future research, where we plan to include such cases to enhance the model’s generalizability. In conclusion, our 3DCRL-Net model demonstrates high accuracy and efficiency in facilitating CRL evaluation of the standard plane, providing reliable CRL measurements comparable to those of radiologists using 2D ultrasound in clinical practice from 11 to 14 weeks. Abbreviations CRL Crown-Rump Length 3DCRL-Net 3D Crown-Rump Length Network 2D-RAD 2D Radiologist Group 3D-RAD 3D Radiologist Group MAE Mean Absolute Error SD Standard Deviation CI Confidence Interval T Test Statistic χ² Chi-Square Statistic P Probability (p-value in statistical tests) ISUOG International Society of Ultrasound in Obstetrics and Gynecology GA Gestational Age Declarations Ethics approval and consent to participate The study was approved by the Medical Research Ethics Committee (Approval No. 2019-LHQRMYY-LL-062). Written informed consent was obtained from all participating pregnant women. The study was conducted ethically in accordance with the World Medical Association Declaration of Helsinki. Consent for Publication Not Applicable. Availability of data and material The datasets and code are not publicly available due to the hospital policy and personal privacy but are available from the corresponding author on reasonable request. Competing interests The authors declare that they have no competing interests. Author Contribution Y.Z., Y.H., and C.C. wrote the main manuscript text and prepared Figures 1-5 and Tables 1-4. X.H. contributed to case collection and the preparation of Figure 3. W.P., H.L., and Y.H. participated in some of the experiments. Y.C. provided tool support, and Y.Y., Y.X., and D.N. supervised the entire experiment. All authors reviewed the manuscript. References International Society of Ultrasound in Obstetrics and Gynecology, Bilardo CM, Chaoui R, Hyett JA, Kagan KO, Karim JN, Papageorghiou AT, Poon LC, Salomon LJ, Syngelaki A, Nicolaides KH. ISUOG Practice Guidelines (updated): performance of 11–14-week ultrasound scan. Ultrasound Obstet Gynecol. 2023;61:127-143. Hadlock FP, Shah YP, Kanon DJ, Lindsey JV. Fetal crown-rump length: reevaluation of relation to menstrual age (5-18 weeks) with high-resolution real-time US. Radiology . 1992;182:501-505. Dhombres F, Khoshnood B, Bessis R, Fries N, Senat MV, Jouannic JM. Quality of first-trimester measurement of crown-rump length. Am J Obstet Gynecol . 2014;211:672.e1-672.e5. Spencer K. Aneuploidy screening in the first trimester. Am J Med Genet Part C: Semin Med Genet. 2007;145C:18-32. Van Heesch PN, Struijk PC, Laudy JAM, Steegers EAP, Wildschut HIJ. Estimating the effect of gestational age on test performance of combined first-trimester screening for Down syndrome: A preliminary study. J Perinat Med . 2010;38. Pandya PP, Altman DG, Brizot ML, Pettersen H, Nicolaides KH. Repeatability of measurement of fetal nuchal translucency thickness. Ultrasound Obstet Gynecol . 1995;5:334-337. Napolitano R, Dhami J, Ohuma E, Ioannou C, Conde‐Agudelo A, Kennedy S, Villar J, Papageorghiou A. Pregnancy dating by fetal crown–rump length: A systematic review of charts. BJOG . 2014;121:556-565. Salomon LJ, Alfirevic Z, Da Silva Costa F, Deter RL, Figueras F, Ghi T, Glanc P, Khalil A, Lee W, Napolitano R, Papageorghiou A, Sotiriadis A, Stirnemann J, Toi A, Yeo G. ISUOG Practice Guidelines: ultrasound assessment of fetal biometry and growth. Ultrasound Obstet Gynecol . 2019;53:715-723. Wanyonyi SZ, Napolitano R, Ohuma EO, Salomon LJ, Papageorghiou AT. Image-scoring system for crown-rump length measurement. Ultrasound Obstet Gynecol . 2014;44:649-654. Zhang Y, Niu Z, Meng H, Jiang Y, Xu Z, Ouyang Y, et al. Gestational age reference from crown-rump length during 11–14 weeks: a population-based multicenter cohort study in China. BMC Pregnancy Childbirth. 2025;25(1):214. Goldstein SR. Embryonic ultrasonographic measurements: crown-rump length revisited. Am J Obstet Gynecol. 1991;165:497-501. Tsai PY, Hung CH, Chen CY, Sun YN. Automatic fetal middle sagittal plane detection in ultrasound using generative adversarial network. Diagnostics. 2020;11:21. Rawat V, Jain A, Shrimali V. Automated techniques for the interpretation of fetal abnormalities: A review. Appl Bionics Biomech . 2018:1-11. Falcon O, Cavoretto P, Peralta CFA, Csapo B, Nicolaides KH. Fetal head-to-trunk volume ratio in chromosomally abnormal fetuses at 11 + 0 to 13 + 6 weeks of gestation. Ultrasound Obstet Gynecol. 2005;26:755-760. Santorum M, Wright D, Syngelaki A, Karagioti N, Nicolaides KH. Accuracy of first-trimester combined test in screening for trisomies 21, 18, and 13. Ultrasound Obstet Gynecol . 2017;49:714-720. Luchi C, Persico N, Rembouskos G, Nicolaides KH. Practical approach to obtain the mid-sagittal plane of the fetal face at 11–13 weeks’ gestation by two-dimensional ultrasound. Ultrasound Obstet Gynecol. 2014;44:617-618. Gjessing HK, Grøttum P, Dreier JM, Eik-Nes SH. Biparietal diameter vs crown-rump length as standard parameter for late first-trimester pregnancy dating. Ultrasound Obstet Gynecol . 2024 Dec;64(6):739-745. doi: 10.1002/uog.29124. Zhao L, Asis-Cruz JD, Feng X, Wu Y, Kapse K, Largent A, et al. Automated 3D fetal brain segmentation using an optimized deep learning approach. Am J Neuroradiol . 2022;43:448-454. Grangé G, Althuser M, Fresson J, Bititi A, Miyamoto K, Tsatsaris V, Morel O. Semi-automated adjusted measurement of nuchal translucency: feasibility and reproducibility. Ultrasound Obstet Gynecol. 2011;37:335-340. Li Y, Khanal B, Hou B, Alansary A, Cerrolaza JJ, Sinclair M, et al. Standard plane detection in 3D fetal ultrasound using an iterative transformation network. In: Frangi AF, Schnabel JA, Davatzikos C, Alberola-López C, Fichtinger G, editors. Medical Image Computing and Computer Assisted Intervention – MICCAI 2018. 2018;11070:392-400. Elmahallawy SEM, Eid SM, Mahmoud MS. Comparative study between crown-rump length and fetal volume in first trimester for accurate estimation of gestational age. Int J Med Arts . 2024;6(2):4163-4168. Liang J, Yang X, Huang Y, Li H, He S, Hu X, et al. Sketch guided and progressive growing GAN for realistic and editable ultrasound image synthesis. Med Image Anal . 2022;79:101-112. Xie B, Lei T, Wang N, Cai H, Xian J, He M, et al. Computer-aided diagnosis for fetal brain ultrasound images using deep convolutional neural networks. Int J Comput Ass Rad . 2020;15:1303-1312. Sofka M, Zhang J, Good S, Zhou SK, Comaniciu D. Automatic detection and measurement of structures in fetal head ultrasound volumes using sequential estimation and integrated detection network (IDN). IEEE Trans Med Imag . 2014;33:1054-1070. Li J, Wang Y, Lei B, Cheng JZ, Qin J, Wang T, et al. Automatic fetal head circumference measurement in ultrasound using random forest and fast ellipse fitting. IEEE J Biomed Health Inform . 2018;22:215-223. Silver D, Schrittwieser J, Simonyan K, Antonoglou I, Huang A, Guez A, et al. Mastering the game of Go without human knowledge. Nature . 2017;550:354-359. Drukker L. Real-time localization of fetal anomalies on ultrasound using artificial intelligence: what’s next? Ultrasound Obstet Gynecol . 2022;59:285-287. Liang, J., Yang, X., Huang, Y., Li, H., He, S., Hu, X., Chen, Z., Xue, W., Cheng, J., Ni, D.: Sketch guided and progressive growing gan for realistic and editable ultrasound image synthesis. Medical Image Analysis 79, 102461 (2022) Chen C, Yang X, Huang Y, Shi W, Cao Y, Luo M, et al. FetusMapV2: Enhanced fetal pose estimation in 3D ultrasound. Med Image Anal . 2024;91:103013. Blaas H ‐G. K, Taipale P, Torp H, Eik‐Nes SH. Three‐dimensional ultrasound volume calculations of human embryos and young fetuses: A study on the volumetry of compound structures and its reproducibility. Ultrasound in Obstet & Gyne . 2006;27:640–6. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 16 Jul, 2025 Read the published version in BMC Pregnancy and Childbirth → Version 1 posted Editorial decision: Revision requested 23 May, 2025 Reviews received at journal 21 May, 2025 Reviewers agreed at journal 05 May, 2025 Reviews received at journal 12 Apr, 2025 Reviewers agreed at journal 12 Apr, 2025 Reviewers invited by journal 29 Mar, 2025 Editor invited by journal 24 Mar, 2025 Editor assigned by journal 21 Mar, 2025 Submission checks completed at journal 21 Mar, 2025 First submitted to journal 18 Mar, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6250944","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":443528656,"identity":"38d02179-aae0-40bd-a623-40bdf4eb0f1e","order_by":0,"name":"Yuanji Zhang","email":"data:image/png;base64,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","orcid":"","institution":"Shenzhen University","correspondingAuthor":true,"prefix":"","firstName":"Yuanji","middleName":"","lastName":"Zhang","suffix":""},{"id":443528657,"identity":"d2603fcb-735d-47d4-8b2f-da954dc1acb4","order_by":1,"name":"Yuhao Huang","email":"","orcid":"","institution":"Shenzhen University","correspondingAuthor":false,"prefix":"","firstName":"Yuhao","middleName":"","lastName":"Huang","suffix":""},{"id":443528658,"identity":"495183df-0487-4562-a4fb-cfb012cbecac","order_by":2,"name":"Chaoyu Chen","email":"","orcid":"","institution":"Shenzhen University","correspondingAuthor":false,"prefix":"","firstName":"Chaoyu","middleName":"","lastName":"Chen","suffix":""},{"id":443528659,"identity":"ee2d25ab-d1ff-4aad-bb95-923794105b2d","order_by":3,"name":"Xing Hu","email":"","orcid":"","institution":"Shenzhen Luohu People's Hospital, Third Affiliated Hospital of Shenzhen University","correspondingAuthor":false,"prefix":"","firstName":"Xing","middleName":"","lastName":"Hu","suffix":""},{"id":443528660,"identity":"ff91bff1-dd2b-4e21-8a8e-ddcd39f149fe","order_by":4,"name":"Wenxiong Pan","email":"","orcid":"","institution":"Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Wenxiong","middleName":"","lastName":"Pan","suffix":""},{"id":443528661,"identity":"005e6c2d-4ef3-41f8-942f-939c7570423b","order_by":5,"name":"Huanjia Luo","email":"","orcid":"","institution":"Huizhou Central People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Huanjia","middleName":"","lastName":"Luo","suffix":""},{"id":443528663,"identity":"8dcdd3ff-29ef-4408-ac09-22702b19320e","order_by":6,"name":"Yankai Huang","email":"","orcid":"","institution":"Southern Medical University Shenzhen Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yankai","middleName":"","lastName":"Huang","suffix":""},{"id":443528665,"identity":"97472ba3-ed48-4775-b79e-9b019b1013cd","order_by":7,"name":"Haixia Wang","email":"","orcid":"","institution":"Shenzhen Luohu People's Hospital, Third Affiliated Hospital of Shenzhen University","correspondingAuthor":false,"prefix":"","firstName":"Haixia","middleName":"","lastName":"Wang","suffix":""},{"id":443528667,"identity":"7ce0ed12-7dcf-4395-9afd-693e8903339c","order_by":8,"name":"Yan Cao","email":"","orcid":"","institution":"Shenzhen RayShape Medical Technology Co., Ltd","correspondingAuthor":false,"prefix":"","firstName":"Yan","middleName":"","lastName":"Cao","suffix":""},{"id":443528669,"identity":"75ff0e29-5702-45c1-9559-82fb36fc40c2","order_by":9,"name":"Yan Yi","email":"","orcid":"","institution":"ShenZhen People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yan","middleName":"","lastName":"Yi","suffix":""},{"id":443528670,"identity":"1560bfa0-2d3d-461b-b2b2-6fcbb8661cc5","order_by":10,"name":"Yi Xiong","email":"","orcid":"","institution":"Shenzhen Luohu People's Hospital, Third Affiliated Hospital of Shenzhen University","correspondingAuthor":false,"prefix":"","firstName":"Yi","middleName":"","lastName":"Xiong","suffix":""},{"id":443528671,"identity":"adc3231a-f0e9-410b-8e88-f3c448b83584","order_by":11,"name":"Dong Ni","email":"","orcid":"","institution":"Shenzhen University","correspondingAuthor":false,"prefix":"","firstName":"Dong","middleName":"","lastName":"Ni","suffix":""}],"badges":[],"createdAt":"2025-03-18 08:23:35","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6250944/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6250944/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12884-025-07823-6","type":"published","date":"2025-07-16T15:57:07+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":81014458,"identity":"ad0224f9-56e2-44be-b7bb-ecddc6b38e41","added_by":"auto","created_at":"2025-04-21 08:40:40","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":139423,"visible":true,"origin":"","legend":"\u003cp\u003eData Collection for Internal and External Datasets.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6250944/v1/51bce589b085989ef92d1d7f.jpg"},{"id":81016374,"identity":"e7041ff8-0570-4453-bb37-075052263682","added_by":"auto","created_at":"2025-04-21 08:56:39","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":116651,"visible":true,"origin":"","legend":"\u003cp\u003eThe Framework of the 3DCRL-Net Model. GCP: gradient checkpoint; OLS: Ordinary Least Squares; REV: reversible layer.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6250944/v1/4acf01537fba00460883bb15.jpg"},{"id":81014455,"identity":"55fee6e2-ec9e-41ea-9b6a-5b7fa28cb6be","added_by":"auto","created_at":"2025-04-21 08:40:39","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":124204,"visible":true,"origin":"","legend":"\u003cp\u003eFetal Cases of Standard and Non-standard CRL Planes Obtained by 3DCRL-Net, 2D-RAD, and 3D-RAD. Non-standard planes failed to meet the required criteria, including the midsagittal plane, neutral position, horizontal orientation, clear visibility of the crown and rump, and adequate magnification.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6250944/v1/a6d08bd2a7bc19dd5d4543c5.jpg"},{"id":81014452,"identity":"18cf208e-a522-4fd0-802f-58dbab9eddc0","added_by":"auto","created_at":"2025-04-21 08:40:39","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":70546,"visible":true,"origin":"","legend":"\u003cp\u003eBland-Altman Plot Showing the Agreement Distribution of CRL Measurements Between 2D-RAD and 3DCRL-Net, and Between 2D-RAD and 3D-RAD. Blue dots represent standard CRL measurements, and orange dots represent non-standard CRL measurements.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6250944/v1/3ec6d81adf71e596f8ad78bb.jpg"},{"id":81014456,"identity":"afd4b47c-b4c2-4749-a35e-2b716479305c","added_by":"auto","created_at":"2025-04-21 08:40:40","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":71042,"visible":true,"origin":"","legend":"\u003cp\u003eBox Plot of CRL Measurement Among the Three Groups at Different Gestational Ages (GAs) from 11 to 14 Weeks.\u003cstrong\u003e \u003c/strong\u003eThe dark blue box represents 3D-RAD, the light blue box represents 2D-RAD, and the yellow box represents 3DCRL-Net. The boxes indicate the standard deviation, the '□' represents the mean, and the '--' represents the median.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6250944/v1/08619f5b5f55e8b9148a9b09.jpg"},{"id":87219357,"identity":"cedc6540-4cc2-405c-96f5-0bee8aa447c0","added_by":"auto","created_at":"2025-07-21 16:04:22","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1543743,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6250944/v1/036097c6-a37c-470f-b1fe-9f01dfed1dfd.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Comparative Study of 2D vs. 3D AI-Enhanced Ultrasound for Accurate Fetal Crown–rump Length Evaluation in the First Trimester","fulltext":[{"header":"1 Background","content":"\u003cp\u003eAccurate fetal growth evaluation is crucial for monitoring fetal health and predicting the risk of anomalies [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The crown-rump length (CRL) is considered the gold standard for estimating gestational age (GA) and assessing fetal growth during the first trimester [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. A precise CRL measurement aids in diagnosing fetal growth restriction, enables accurate evaluation of nuchal translucency (NT) for assessing the risk of aneuploidies, and enhances the detection of structural anomalies [\u003cspan additionalcitationids=\"CR5 CR6 CR7 CR8\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe standard CRL plane and measurements must meet strict quality control standards to ensure diagnostic accuracy, but CRL evaluation depends heavily on the radiologist's experience [\u003cspan additionalcitationids=\"CR11 CR12 CR13 CR14\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Identifying the optimal fetal position and acquiring the correct plane during two-dimensional (2D) screening is time-consuming and prone to errors [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Significant variability between radiologists can further lead to diagnostic inaccuracies [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. While three-dimensional (3D) ultrasound allows offline manipulation for accurate CRL measurement, it remains challenging due to the complexity of data processing and the need for familiarity with the system [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eArtificial intelligence (AI) has demonstrated potential in predicting GA from 3D data in the first trimester [\u003cspan additionalcitationids=\"CR23\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. However, limited AI validation in CRL analysis, including plane localization and measurement accuracy, comparison with traditional methods (e.g., 2D screening), and improved interpretability, hampers clinician trust and adoption of DL-based systems [\u003cspan additionalcitationids=\"CR26\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. In this study, we developed an AI-based model using the 3D U-Net (named 3DCRL-Net). It accurately detects multiple landmarks to regress the standard plane and measure CRL for assessing fetal growth during the first trimester. We also compared its performance with experienced radiologists using both 2D and 3D ultrasound screening to explore its potential in clinical practice.\u003c/p\u003e"},{"header":"2 Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Study design and data collection\u003c/h2\u003e \u003cp\u003eIn this prospective cohort study, fetal data, including 2D videos and 3D volumes, were collected from 1326 fetal ultrasound screenings conducted between 11 and 14 weeks of gestation at the Third Affiliated Hospital of Shenzhen University, from June 2021 to June 2023. The study was approved by the Medical Research Ethics Committee (Approval No. 2019-LHQRMYY-LL-062). Written informed consent was obtained from all participating pregnant women. The inclusion criteria for the data were singleton fetuses with a CRL between 45 mm and 84 mm, without structural abnormalities, and a fetal position conducive to obtaining CRL plane.\u003c/p\u003e \u003cp\u003eAll ultrasound examinations were performed using the following equipment: General Electric Voluson E6/E8/E10 (GE Healthcare, Zipf, Austria), Philips EPIQ7 (Philips, Seattle, Washington, USA), and Mindray Resona-7 (Mindray, Shenzhen, Guangdong Province, China), with a 3D transabdominal volume probe. The standard CRL plane should meet the five criteria outlined by the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]: midsagittal plane, neutral position, horizontal orientation (crown-rump line at 75\u0026deg; to 105\u0026deg;), clear visibility of the crown and rump, and adequate magnification (over two-thirds of the image).\u003c/p\u003e \u003cp\u003eFor each fetus, the radiologists adjusted the volume box to encompass the crown and rump for 3D acquisition and ensured proper adjustment of 2D ultrasound videos to obtain the standard CRL images. Cases without optimal CRL within 15 minutes were excluded, and successful images and volumes were saved for analysis. The time cost for both 2D video and 3D volume was recorded. Quality control was performed by three radiology specialists with 20\u0026ndash;30 years of experience in fetal examination to ensure the presence of the standard CRL plane in all fetal data. Only data confirmed by at least two radiologists were included in the study. The data collection process is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Annotation of internal and external dataset\u003c/h2\u003e \u003cp\u003eIn the internal dataset, a total of 1,000 fetal volumes were collected and divided into training, validation, and testing sets in an 8:1:1 ratio. For CRL analysis and model development, each case was annotated with nine landmarks, a standard CRL plane, and a CRL measurement. The nine structural landmarks included the cranial crest, thalamus, diencephalon, nasal bone, lower alveolar, chest wall, diaphragm, lumbar, and buttocks. Besides, the standard CRL plane equation (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ax+by+cz+d=0\\)\u003c/span\u003e\u003c/span\u003e) and CRL measurement were also annotated for model evaluation. This process was facilitated using the Pair Annotation Software Package 3.0 [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] by three experienced radiologists (\u0026gt;\u0026thinsp;5 years of fetal examination experience). Three radiology specialists with 20\u0026ndash;30 years of experience performed quality control on the annotations made by the experienced radiologists.\u003c/p\u003e \u003cp\u003eIn the external dataset, 245 fetal cases, including both 2D and 3D data, were collected. To compare the performance of 3DCRL-Net vs. radiologists, the same three radiologists independently localized the standard CRL plane in both the 2D and 3D data, measured the CRL, and recorded the time taken to identify the plane and measure the CRL for each case.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 The framework of 3DCRL-Net Model\u003c/h2\u003e \u003cp\u003eThe framework of 3DCRL-Net model was shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Building on our previous work in 3D fetal ultrasound anatomical landmark detection (FetusMapV2 [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]), we developed a simplified version for CRL analysis. The proposed framework mainly includes a c landmark detector, taking the 3D fetus as input and outputting \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e CRL-related landmarks with Gaussian heatmap representation. During training, we leveraged the structure-unconstrained gradient checkpoint (GCP) and activation-unreserve reversible layer (REV) techniques to reduce the memory requirements and enable the low-resource training on limited GPU memory. This supports high-resolution input, finally enhancing the landmark detection accuracy. The Kullback-Leibler (KL) divergency loss was used to optimize the learning process by calculating the distance between annotated and predicted heatmap distributions.\u003c/p\u003e \u003cp\u003eDuring testing, the detector takes the entire 3D volume as input and generates Gaussian heatmaps of landmarks as output. Then, the position with the highest value of each heatmap will be considered as the final landmark prediction (i.e., [x, y, z]). Based on the detected landmarks:\u003c/p\u003e \u003cp\u003e(a) We used the Ordinary Least Squares (OLS) method to regress the plane equation. The goal is to minimize the sum of squared Euclidean distances from these landmarks to the plane. First, we calculate the center of these landmarks (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\stackrel{-}{x},\\stackrel{-}{y},\\stackrel{-}{z}\\)\u003c/span\u003e\u003c/span\u003e) and decentralize them to eliminate positional influence, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{r}_{i}=({x}_{i}-\\stackrel{-}{x},{y}_{i}-\\stackrel{-}{y},{z}_{i}-\\stackrel{-}{z})\\)\u003c/span\u003e\u003c/span\u003e. Next, we construct the covariance matrix \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:(U=\\sum\\:{r}_{j}{r}_{j}^{T})\\)\u003c/span\u003e\u003c/span\u003e and perform eigenvalue decomposition (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Uv=\\lambda\\:v\\)\u003c/span\u003e\u003c/span\u003e). We then select the eigenvector \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{v}_{min}=(a,b,c)\\)\u003c/span\u003e\u003c/span\u003e corresponding to the smallest eigenvalue as the plane's normal vector. Given that the plane passes through the center point, the offset \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:d\\)\u003c/span\u003e\u003c/span\u003e of the plane equation is calculated as: \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:d=-(a\\stackrel{-}{x}+b\\stackrel{-}{y}+c\\stackrel{-}{z}))\\)\u003c/span\u003e\u003c/span\u003e. Finally, the CRL plane equation in 3D space can be defined as:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:ax+by+cz+d=0.$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003e(b) We measure the CRL by calculating the distance between two landmarks (cranial crest and buttocks).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Experiments on Internal and External Testing Datasets\u003c/h2\u003e \u003cp\u003eWe implemented our proposed landmark detector in Python and Pytorch, using one NVIDIA GTX 2080Ti GPU with 12 GB memory. All volumes were resized to 208*128*144 to maximize the memory usage. We used Adam optimizer with learning rate of 1e-4 and moment term of 0.5, to update the model weights. The total training epoch is 100 and batch size is 1.\u003c/p\u003e \u003cp\u003eFor the standard CRL plane evaluation, three radiology specialists with 20\u0026ndash;30 years of experience audited the accuracy of CRL plane localization for the 3DCRL-Net, 2D-radiologist (2D-RAD), and 3D-radiologist (3D-RAD) groups. Each image was assessed and assigned a binary score: 0 (non-standard plane) and 1 (standard plane). Only results where at least two specialists agreed on the same score (either 0 or 1) were considered valid.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003e1) The Performance of 3DCRL-Net in Internal Testing Datasets\u003c/h3\u003e\n\u003cp\u003eThe performance of the 3DCRL-Net model in the internal testing dataset was evaluated for CRL landmarks, plane localization, and measurement. The 3DCRL-Net model's results were compared with 3D-RAD for the nine key landmarks in distance error assessment, CRL plane localization accuracy using both subjective evaluations by specialists and objective comparisons, and CRL measurement results on the standard plane. Time efficiency was also compared between the two groups.\u003c/p\u003e\n\u003ch3\u003e2) Accuracy of 3DCRL-Net in External Testing Datasets\u003c/h3\u003e\n\u003cp\u003eThe external dataset was used to validate the accuracy and efficiency of the 3DCRL-Net model for CRL plane localization and measurements. The performance of 3DCRL-Net in automatically identifying and measuring CRL was compared with 2D-RAD and 3D-RAD, audited by three specialists. Time efficiency was also compared across the radiologist groups.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Statistical analysis\u003c/h2\u003e \u003cp\u003eTo assess the performance of CRL structural landmarks, plane localization and measurement, several metrics were employed, including mean, mean absolute error, mean difference, standard deviation (SD), and confidence intervals (CI). Additionally, plane angular deviation and landmark distance errors were calculated. Interobserver variability was assessed using Bland-Altman plots. Differences in measurement techniques were analyzed using chi-square tests and T-tests. Time costs were also evaluated. All statistical analyses were conducted using Python 3.11 (Python Software Foundation), with a significance threshold of \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Characteristics of the study\u003c/h2\u003e \u003cp\u003eThe internal training and validation dataset consisted of 748 and 126 fetal volumes, respectively, the internal testing dataset contained 126 fetal volumes, and the external dataset comprised 245 fetal videos and volumes. The average gestational age of the fetuses in the internal testing dataset and external dataset was 12\u0026thinsp;\u0026plusmn;\u0026thinsp;3 weeks, and 12\u0026thinsp;\u0026plusmn;\u0026thinsp;2 weeks, respectively.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Performance of 3DCRL-Net model on internal testing dataset\u003c/h2\u003e \u003cdiv id=\"Sec13\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Assessment of structural landmarks\u003c/h2\u003e \u003cp\u003eThe performance of the 3DCRL-Net model in predicting landmarks was evaluated by comparing its results with 3D-RAD, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The average absolute distance error for the nine landmarks was 1.81mm, with a SD of 1.10mm. Specifically, the error for the cranial-crest was 1.68\u0026thinsp;\u0026plusmn;\u0026thinsp;0.82 mm, and for the buttocks, it was 2.02\u0026thinsp;\u0026plusmn;\u0026thinsp;2.06mm. These results indicate that the 3DCRL-Net model can accurately identify the cranial-crest and buttocks landmarks, laying a solid foundation for measuring CRL in 3D volume.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of Internal Testing Dataset Experimental Results for Nine Landmarks Localization\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStructural Landmark\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDistance Error (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD) (mm)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCranial crest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.68\u0026thinsp;\u0026plusmn;\u0026thinsp;0.82\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThalamus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.54\u0026thinsp;\u0026plusmn;\u0026thinsp;0.69\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiencephalon\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.63\u0026thinsp;\u0026plusmn;\u0026thinsp;0.82\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNasal bone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.54\u0026thinsp;\u0026plusmn;\u0026thinsp;0.99\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLower alveolar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.50\u0026thinsp;\u0026plusmn;\u0026thinsp;0.71\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChest wall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.02\u0026thinsp;\u0026plusmn;\u0026thinsp;1.11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiaphragm lumbar\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.72\u0026thinsp;\u0026plusmn;\u0026thinsp;0.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUmbilical\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.62\u0026thinsp;\u0026plusmn;\u0026thinsp;1.79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eButtocks\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e2.02\u0026thinsp;\u0026plusmn;\u0026thinsp;2.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e1.81\u0026thinsp;\u0026plusmn;\u0026thinsp;1.10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Objective and subjective assessment of CRL plane\u003c/h2\u003e \u003cp\u003eTo assess the accuracy of the predicted CRL planes derived from landmark regression, we conducted both objective and subjective evaluations on 126 cases, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The objective evaluation measured the spatial differences between the 3DCRL-Net and 3D-RAD. Specifically, we computed the angular deviation with a precision of 5.34\u0026deg; to quantify the discrepancy between the two planes. For the subjective evaluation audited by specialists, the 3DCRL-Net model achieved an accuracy of 91.27% (115 standard CRL planes), while the 3D-RAD had an accuracy of 80.16% (101 cases met the criteria).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e3.2.3 Assessment of CRL measurement\u003c/h2\u003e \u003cp\u003eThe performance of 3DCRL-Net in measuring CRL on the standard plane is presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The CRL measurement in 3D is defined as the distance between the cranial-crest and the rump landmarks. The results show that the mean and SD of CRL measurements obtained by 3DCRL-Net and 3D-RAD are 63.06mm\u0026thinsp;\u0026plusmn;\u0026thinsp;8.13mm and 62.60mm\u0026thinsp;\u0026plusmn;\u0026thinsp;7.93mm, respectively, with no significant difference between them (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.70). Moreover, due to the 3DCRL-Net 's accurate prediction of the cranial-crest and rump landmarks, the CRL measurement estimated by the 3DCRL-Net closely matches the 3D-RAD results, with a mean absolute error of 1.26mm\u0026thinsp;\u003cb\u003e\u0026plusmn;\u003c/b\u003e\u0026thinsp;1.14mm.\u003c/p\u003e \u003cp\u003eWe recorded the total time from inputting the 3D volume to outputting the CRL measurement for both 3DCRL-Net and 3D-RAD. The mean time cost for fetal cases was 2.02 seconds for 3DCRL-Net and 10 minutes for 3D-RAD (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eObjective and Subjective Evaluations of CRL Plane Localization and Measurement Results\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCRL plane\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAngle Error (\u0026deg;)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3DCRL-Net Accuracy (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3D-RAD Accuracy (%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e91.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e80.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCRL measurement\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeasurement Mean Absolute Error\u003c/p\u003e \u003cp\u003e(mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD) (mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3DCRL-Net Measurement\u003c/p\u003e \u003cp\u003e(mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3D-RAD Measurement\u003c/p\u003e \u003cp\u003e(mm)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1.26\u0026thinsp;\u0026plusmn;\u0026thinsp;1.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e63.06\u0026thinsp;\u0026plusmn;\u0026thinsp;8.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e62.60\u0026thinsp;\u0026plusmn;\u0026thinsp;7.93\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Performance of 3DCRL-Net model on external testing dataset\u003c/h2\u003e \u003cdiv id=\"Sec17\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1 Comparative Performance in CRL Plane Localization: 3DCRL-Net vs. 2D-RAD vs. 3D-RAD\u003c/h2\u003e \u003cp\u003eFor CRL plane localization, the comparative results from the specialists' audit of the 3DCRL-Net, 2D-RAD, and 3D-RAD groups, including both standard and non-standard planes, are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The 3DCRL-Net model achieved an accuracy of 91.43%, identifying 224 standard CRL planes out of 245 fetal volumes, while the 2D-RAD group identified 228 standard planes (93.06% accuracy), and the 3D-RAD group identified 211 standard planes (86.12% accuracy). The most common reason for a non-standard CRL plane in the 3DCRL-Net group was \"Clear Visibility of Crown and Rump\" whereas in the 2D-RAD of \"Midsagittal Plane\" and 3D-RAD groups, it was \"Clear Visibility of Crown and Rump\". The examples of standard and non-standard CRL planes obtained by the three groups are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAccuracy of CRL Plane Localizations by 3DCRL-Net and Radiologists on the External Testing Dataset. The accuracy of ultrasound (US) images was assessed by radiology specialists. A standard image met all five criteria, while a non-standard image was assigned specific reasons.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eImage Result\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eNo. of Cases\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3DCRL-Net\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2D-RAD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3D-RAD\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStandard\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e224 (91.43%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e228 (93.06%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e211(86.12%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-standard\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21 (8.57%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17 (6.94%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34 (13.88%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReason\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMidsagittal Plane\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNeutral Position\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHorizontal Orientation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClear Visibility of Crown and Rump\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdequate Magnification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2 Comparative Performance in CRL measurement: 3DCRL-Net vs. 2D-RAD vs. 3D-RAD\u003c/h2\u003e \u003cp\u003eFor CRL measurement, the mean and SD of the standard CRL measurements for the 3DCRL-Net, 2D-RAD, and 3D-RAD groups were 59.90 mm\u0026thinsp;\u0026plusmn;\u0026thinsp;6.93mm, 61.05 mm\u0026thinsp;\u0026plusmn;\u0026thinsp;7.51mm, and 60.02 mm\u0026thinsp;\u0026plusmn;\u0026thinsp;7.40mm, respectively. The mean absolute error of CRL measurement between 2D-RAD and 3DCRL-Net was 1.58 mm\u0026thinsp;\u0026plusmn;\u0026thinsp;1.43 mm, and between 2D-RAD and 3D-RAD was 1.37 mm\u0026thinsp;\u0026plusmn;\u0026thinsp;0.95 mm, with no significant differences observed between 2D-RAD and 3DCRL-Net (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.25), or between 2D-RAD and 3D-RAD (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.19, ANOVA). The agreement distribution of CRL measurements between the three groups is shown in the Bland-Altman plot (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). The mean differences and SD between 2D-RAD and 3DCRL-Net, 2D-RAD and 3D-RAD are 1.12 mm\u0026thinsp;\u0026plusmn;\u0026thinsp;3.57mm, and 1.06 mm\u0026thinsp;\u0026plusmn;\u0026thinsp;2.66mm, respectively. Moreover, the mean, median, and 95% CI results across different weeks (11\u0026ndash;14 weeks) are shown in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eWe recorded the total time from scanning the 3D volume to outputting CRL measurements for 3DCRL-Net and 3D-RAD, and from scanning 2D screening to measuring CRL for 2D-RAD. The mean time cost was 5 seconds for 3DCRL-Net (1.02 seconds for model prediction), 11 minutes for 3D-RAD, and 35 seconds for 2D-RAD (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe Mean, Median, and 95% CI of Standard CRL Measurements for the 3DCRL-Net, 2D-RAD, and 3D-RAD Groups Across Weeks 11\u0026ndash;14. Since there was only one case at 14\u0026thinsp;+\u0026thinsp;0 weeks, it was included in the 13-week group for statistical purposes.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroups\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3DCRL-Net\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2D-RAD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3D-RAD\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003e11 Weeks\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean(mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e49.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e50.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e49.21\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedian(mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e49.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e50.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e49.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e95%CI (mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e[48.98,\u0026nbsp;50.42]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[49.57,\u0026nbsp;50.90]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e[48.40,\u0026nbsp;50.02]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e12 Weeks\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean(mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e58.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e59.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e58.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedian(mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e59.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e60.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e59.45\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e95%CI (mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e[58.38,\u0026nbsp;59.54]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[59.29,\u0026nbsp;60.41]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e[58.33, 59.51]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e13 Weeks\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean(mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e70.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e72.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e71.56\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedian(mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e70.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e71.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e71.35\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e95%CI (mm)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e[69.14,\u0026nbsp;71.79]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[71.45,\u0026nbsp;74.45]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e[70.22,\u0026nbsp;72.90]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eIn this study, we developed an AI-based model, named 3DCRL-Net, to acquire the standard CRL plane and accurately measure CRL using 3D volume data during 11 to 14 weeks. The 3DCRL-Net demonstrated high accuracy in identifying structural landmarks (mean absolute distance error, 1.81mm), locating the standard CRL plane (accuracy, 91.27%), and measuring CRL (mean absolute error, 1.26mm) in the internal testing dataset, audited by specialists. In a comparative experiment with experienced radiologists using 2D and 3D ultrasound on the external dataset, 3DCRL-Net (91.43%) achieved comparable accuracy in CRL plane localization to 2D-RAD (93.06%), slightly higher than 3D-RAD (86.12%). The model also showed minimal differences in CRL measurement compared to radiologists using 2D ultrasound, with a mean absolute error of 1.58mm and mean difference of 1.12 mm. These results suggest that 3DCRL-Net has strong potential for fetal crown-rump length evaluation in the first trimester.\u003c/p\u003e \u003cp\u003eAccurate CRL measurement is essential for fetal growth assessment and clinical management [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. In practice, it relies on radiologists' ability to precisely locate the standard plane, demanding extensive knowledge and skill [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Previous studies have highlighted the global challenge of significant variability among radiologists, leading to inaccurate CRL plane acquisition and measurement [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan additionalcitationids=\"CR11 CR12\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Traditional CRL plane localization and measurement in 2D screening are prone to errors and are time-consuming for obtaining the standard plane. In 3D volumes, expertise is still required to handle high-dimensional data and spatial awareness, presenting an ongoing challenge [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Therefore, this study developed the 3DCRL-Net to automatically detect anatomical structures in the 3D fetal volume, minimizing the technical expertise required from the radiologist. It located the standard CRL plane and performed measurements based on the detected landmarks. This model provided reliable and efficient results, reducing operator-related variability and improving the accuracy of fetal growth evaluation.\u003c/p\u003e \u003cp\u003eCompared to 2D ultrasound, the enhanced volumetric data from 3D ultrasound significantly improves the detection of key anatomical landmarks, aiding in the localization of the standard CRL plane and enhancing the accuracy of CRL measurements for fetal growth evaluation [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. In our 3DCRL-Net, we determine the standard CRL plane by detecting nine key structural landmarks on the midsagittal plane, rather than relying solely on single-plane localization within the 3D volume. This approach helps mitigate biases and measurement variability caused by subjective radiologist interpretation of plane identification. In the task of detecting the nine landmarks, the 3DCRL-Net model demonstrated high consistency with experienced radiologists using 3D ultrasound, indicating its strong performance in landmark detection. For measuring CRL at two key anatomical points, the mean absolute distance for the cranial crest and buttocks in 3D volume was less than 2 mm.\u003c/p\u003e \u003cp\u003eFor CRL plane localization assessment, our 3DCRL-Net model demonstrated high accuracy, comparable to experienced radiologists using both 2D and 3D ultrasound in standard CRL plane localization. In the external dataset, 3DCRL-Net achieved performance comparable to 2D-RAD and outperformed 3D-RAD, with a 5.31% improvement in accuracy. This highlights the 3DCRL-Net model's effectiveness in locating the standard CRL plane in 3D volumes through key structural landmark detection, making it suitable for clinical practice. Radiologists were more likely to fail in identifying the midsagittal plane with 2D ultrasound, and both the midsagittal plane and crown-rump visibility with 3D ultrasound. In contrast, the 3DCRL-Net model mainly struggled with crown-rump visibility. These challenges highlight the difficulty in accurately identifying the midsagittal plane. Fetal posture issues during 3D volume scanning, such as crown-rump visibility, neutral position, and horizontal orientation, can lead to non-standard CRL plane localization.\u003c/p\u003e \u003cp\u003eFor the comparative results of CRL measurement in the standard CRL plane, the 3DCRL-Net demonstrated high consistency with radiologists using 2D ultrasound, with mean differences of 1.12mm. Furthermore, in comparison box plots of 3DCRL-Net, 2D-RAD, and 3D-RAD at 11, 12, and 13 weeks, both the mean and median CRL measurements showed high consistency, demonstrating the potential of 3DCRL-Net to accurately identify key anatomical landmarks and measure CRL across different gestational weeks. The time results further highlighted the great potential of 3DCRL-Net in clinical settings, providing precise CRL measurements while saving time in fetal screening\u0026mdash;saving 0.5 minutes compared to 2D-RAD and 10 minutes compared to 3D-RAD. This could potentially transform clinical scanning practices by alleviating the high technical demands on radiologists.\u003c/p\u003e \u003cp\u003eAlthough the 3DCRL-Net model automatically locates the CRL plane and measures CRL with high performance, there are still limitations to consider. First, we did not evaluate the impact of different clinicians' techniques on 3D volume collection, which may require proper training of radiologists to ensure correct fetal positioning. Additionally, we did not assess the model's performance in difficult clinical cases, such as maternal conditions or fetal abnormalities. These aspects will be addressed in future research, where we plan to include such cases to enhance the model\u0026rsquo;s generalizability.\u003c/p\u003e \u003cp\u003eIn conclusion, our 3DCRL-Net model demonstrates high accuracy and efficiency in facilitating CRL evaluation of the standard plane, providing reliable CRL measurements comparable to those of radiologists using 2D ultrasound in clinical practice from 11 to 14 weeks.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eCRL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eCrown-Rump Length\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e3DCRL-Net\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003e3D Crown-Rump Length Network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e2D-RAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003e2D Radiologist Group\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e3D-RAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003e3D Radiologist Group\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eMAE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eMean Absolute Error\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eStandard Deviation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eConfidence Interval\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eTest Statistic\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003e\u0026chi;\u0026sup2;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eChi-Square Statistic\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eProbability (p-value in statistical tests)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eISUOG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eInternational Society of Ultrasound in Obstetrics and Gynecology\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 113px;\"\u003e\n \u003cp\u003eGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 440px;\"\u003e\n \u003cp\u003eGestational Age\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was approved by the Medical Research Ethics Committee (Approval No. 2019-LHQRMYY-LL-062). Written informed consent was obtained from all participating pregnant women. The study was conducted ethically in accordance with the World Medical Association Declaration of Helsinki.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for Publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot Applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and material\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets and code are not publicly available due to the hospital policy and personal privacy but are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eY.Z., Y.H., and C.C. wrote the main manuscript text and prepared Figures 1-5 and Tables 1-4. X.H. contributed to case collection and the preparation of Figure 3. W.P., H.L., and Y.H. participated in some of the experiments. Y.C. provided tool support, and Y.Y., Y.X., and D.N. supervised the entire experiment. All authors reviewed the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eInternational Society of Ultrasound in Obstetrics and Gynecology, Bilardo CM, Chaoui R, Hyett JA, Kagan KO, Karim JN, Papageorghiou AT, Poon LC, Salomon LJ, Syngelaki A, Nicolaides KH. ISUOG Practice Guidelines (updated): performance of 11\u0026ndash;14-week ultrasound scan. \u003cem\u003eUltrasound Obstet Gynecol.\u003c/em\u003e 2023;61:127-143.\u003c/li\u003e\n\u003cli\u003eHadlock FP, Shah YP, Kanon DJ, Lindsey JV. Fetal crown-rump length: reevaluation of relation to menstrual age (5-18 weeks) with high-resolution real-time US. \u003cem\u003eRadiology\u003c/em\u003e. 1992;182:501-505.\u003c/li\u003e\n\u003cli\u003eDhombres F, Khoshnood B, Bessis R, Fries N, Senat MV, Jouannic JM. Quality of first-trimester measurement of crown-rump length. \u003cem\u003eAm J Obstet Gynecol\u003c/em\u003e. 2014;211:672.e1-672.e5.\u003c/li\u003e\n\u003cli\u003eSpencer K. Aneuploidy screening in the first trimester. \u003cem\u003eAm J Med Genet Part C: Semin Med Genet.\u003c/em\u003e 2007;145C:18-32.\u003c/li\u003e\n\u003cli\u003eVan Heesch PN, Struijk PC, Laudy JAM, Steegers EAP, Wildschut HIJ. Estimating the effect of gestational age on test performance of combined first-trimester screening for Down syndrome: A preliminary study. \u003cem\u003eJ Perinat Med\u003c/em\u003e. 2010;38.\u003c/li\u003e\n\u003cli\u003ePandya PP, Altman DG, Brizot ML, Pettersen H, Nicolaides KH. Repeatability of measurement of fetal nuchal translucency thickness. \u003cem\u003eUltrasound Obstet Gynecol\u003c/em\u003e. 1995;5:334-337.\u003c/li\u003e\n\u003cli\u003eNapolitano R, Dhami J, Ohuma E, Ioannou C, Conde‐Agudelo A, Kennedy S, Villar J, Papageorghiou A. Pregnancy dating by fetal crown\u0026ndash;rump length: A systematic review of charts.\u003cem\u003e BJOG\u003c/em\u003e. 2014;121:556-565.\u003c/li\u003e\n\u003cli\u003eSalomon LJ, Alfirevic Z, Da Silva Costa F, Deter RL, Figueras F, Ghi T, Glanc P, Khalil A, Lee W, Napolitano R, Papageorghiou A, Sotiriadis A, Stirnemann J, Toi A, Yeo G. ISUOG Practice Guidelines: ultrasound assessment of fetal biometry and growth. \u003cem\u003eUltrasound Obstet Gynecol\u003c/em\u003e. 2019;53:715-723.\u003c/li\u003e\n\u003cli\u003eWanyonyi SZ, Napolitano R, Ohuma EO, Salomon LJ, Papageorghiou AT. Image-scoring system for crown-rump length measurement. \u003cem\u003eUltrasound Obstet Gynecol\u003c/em\u003e. 2014;44:649-654.\u003c/li\u003e\n\u003cli\u003eZhang Y, Niu Z, Meng H, Jiang Y, Xu Z, Ouyang Y, et al. Gestational age reference from crown-rump length during 11\u0026ndash;14 weeks: a population-based multicenter cohort study in China. \u003cem\u003eBMC Pregnancy Childbirth.\u003c/em\u003e 2025;25(1):214.\u003c/li\u003e\n\u003cli\u003eGoldstein SR. Embryonic ultrasonographic measurements: crown-rump length revisited. \u003cem\u003eAm J Obstet Gynecol.\u003c/em\u003e 1991;165:497-501.\u003c/li\u003e\n\u003cli\u003eTsai PY, Hung CH, Chen CY, Sun YN. Automatic fetal middle sagittal plane detection in ultrasound using generative adversarial network. \u003cem\u003eDiagnostics. \u003c/em\u003e2020;11:21.\u003c/li\u003e\n\u003cli\u003eRawat V, Jain A, Shrimali V. Automated techniques for the interpretation of fetal abnormalities: A review. \u003cem\u003eAppl Bionics Biomech\u003c/em\u003e. 2018:1-11.\u003c/li\u003e\n\u003cli\u003eFalcon O, Cavoretto P, Peralta CFA, Csapo B, Nicolaides KH. Fetal head-to-trunk volume ratio in chromosomally abnormal fetuses at 11 + 0 to 13 + 6 weeks of gestation. \u003cem\u003eUltrasound Obstet Gynecol. \u003c/em\u003e2005;26:755-760.\u003c/li\u003e\n\u003cli\u003eSantorum M, Wright D, Syngelaki A, Karagioti N, Nicolaides KH. Accuracy of first-trimester combined test in screening for trisomies 21, 18, and 13. \u003cem\u003eUltrasound Obstet Gynecol\u003c/em\u003e. 2017;49:714-720.\u003c/li\u003e\n\u003cli\u003eLuchi C, Persico N, Rembouskos G, Nicolaides KH. Practical approach to obtain the mid-sagittal plane of the fetal face at 11\u0026ndash;13 weeks\u0026rsquo; gestation by two-dimensional ultrasound. \u003cem\u003eUltrasound Obstet Gynecol. \u003c/em\u003e2014;44:617-618.\u003c/li\u003e\n\u003cli\u003eGjessing HK, Gr\u0026oslash;ttum P, Dreier JM, Eik-Nes SH. Biparietal diameter vs crown-rump length as standard parameter for late first-trimester pregnancy dating. \u003cem\u003eUltrasound Obstet Gynecol\u003c/em\u003e. 2024 Dec;64(6):739-745. doi: 10.1002/uog.29124.\u003c/li\u003e\n\u003cli\u003eZhao L, Asis-Cruz JD, Feng X, Wu Y, Kapse K, Largent A, et al. Automated 3D fetal brain segmentation using an optimized deep learning approach. \u003cem\u003eAm J Neuroradiol\u003c/em\u003e. 2022;43:448-454.\u003c/li\u003e\n\u003cli\u003eGrang\u0026eacute; G, Althuser M, Fresson J, Bititi A, Miyamoto K, Tsatsaris V, Morel O. Semi-automated adjusted measurement of nuchal translucency: feasibility and reproducibility. \u003cem\u003eUltrasound Obstet Gynecol. \u003c/em\u003e2011;37:335-340.\u003c/li\u003e\n\u003cli\u003eLi Y, Khanal B, Hou B, Alansary A, Cerrolaza JJ, Sinclair M, et al. Standard plane detection in 3D fetal ultrasound using an iterative transformation network. In: Frangi AF, Schnabel JA, Davatzikos C, Alberola-L\u0026oacute;pez C, Fichtinger G, editors. Medical Image Computing and Computer Assisted Intervention \u0026ndash; \u003cem\u003eMICCAI\u003c/em\u003e 2018. 2018;11070:392-400.\u003c/li\u003e\n\u003cli\u003eElmahallawy SEM, Eid SM, Mahmoud MS. Comparative study between crown-rump length and fetal volume in first trimester for accurate estimation of gestational age. \u003cem\u003eInt J Med Arts\u003c/em\u003e. 2024;6(2):4163-4168.\u003c/li\u003e\n\u003cli\u003eLiang J, Yang X, Huang Y, Li H, He S, Hu X, et al. Sketch guided and progressive growing GAN for realistic and editable ultrasound image synthesis. \u003cem\u003eMed Image Anal\u003c/em\u003e. 2022;79:101-112.\u003c/li\u003e\n\u003cli\u003eXie B, Lei T, Wang N, Cai H, Xian J, He M, et al. Computer-aided diagnosis for fetal brain ultrasound images using deep convolutional neural networks. \u003cem\u003eInt J Comput Ass Rad\u003c/em\u003e. 2020;15:1303-1312.\u003c/li\u003e\n\u003cli\u003eSofka M, Zhang J, Good S, Zhou SK, Comaniciu D. Automatic detection and measurement of structures in fetal head ultrasound volumes using sequential estimation and integrated detection network (IDN). \u003cem\u003eIEEE Trans Med Imag\u003c/em\u003e. 2014;33:1054-1070.\u003c/li\u003e\n\u003cli\u003eLi J, Wang Y, Lei B, Cheng JZ, Qin J, Wang T, et al. Automatic fetal head circumference measurement in ultrasound using random forest and fast ellipse fitting. \u003cem\u003eIEEE J Biomed Health Inform\u003c/em\u003e. 2018;22:215-223.\u003c/li\u003e\n\u003cli\u003eSilver D, Schrittwieser J, Simonyan K, Antonoglou I, Huang A, Guez A, et al. Mastering the game of Go without human knowledge. \u003cem\u003eNature\u003c/em\u003e. 2017;550:354-359.\u003c/li\u003e\n\u003cli\u003eDrukker L. Real-time localization of fetal anomalies on ultrasound using artificial intelligence: what\u0026rsquo;s next? \u003cem\u003eUltrasound Obstet Gynecol\u003c/em\u003e. 2022;59:285-287.\u003c/li\u003e\n\u003cli\u003eLiang, J., Yang, X., Huang, Y., Li, H., He, S., Hu, X., Chen, Z., Xue, W., Cheng, J., Ni, D.: Sketch guided and progressive growing gan for realistic and editable ultrasound image synthesis. \u003cem\u003eMedical Image Analysis \u003c/em\u003e79, 102461 (2022)\u003c/li\u003e\n\u003cli\u003eChen C, Yang X, Huang Y, Shi W, Cao Y, Luo M, et al. FetusMapV2: Enhanced fetal pose estimation in 3D ultrasound. \u003cem\u003eMed Image Anal\u003c/em\u003e. 2024;91:103013.\u003c/li\u003e\n\u003cli\u003eBlaas H ‐G. K, Taipale P, Torp H, Eik‐Nes SH. Three‐dimensional ultrasound volume calculations of human embryos and young fetuses: A study on the volumetry of compound structures and its reproducibility. \u003cem\u003eUltrasound in Obstet \u0026amp; Gyne\u003c/em\u003e. 2006;27:640\u0026ndash;6.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-pregnancy-and-childbirth","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"prch","sideBox":"Learn more about [BMC Pregnancy and Childbirth](http://bmcpregnancychildbirth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/prch/default.aspx","title":"BMC Pregnancy and Childbirth","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Crown-Rump Length, CRL Evaluation, Artificial Intelligence, Fetal Growth","lastPublishedDoi":"10.21203/rs.3.rs-6250944/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6250944/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eAccurate fetal growth evaluation is crucial for monitoring fetal health, with crown-rump length (CRL) being the gold standard for estimating gestational age and assessing growth during the first trimester. To enhance CRL evaluation accuracy and efficiency, we developed an AI-based model (3DCRL-Net) using the 3D U-Net architecture for automatic landmark detection to achieve CRL plane localization and measurement in 3D ultrasound. We then compared its performance to that of experienced radiologists using both 2D and 3D ultrasound for fetal growth assessment.\u003c/p\u003e\u003ch2\u003eMaterials and methods\u003c/h2\u003e \u003cp\u003eThis prospective cohort study collected fetal data from 1,326 ultrasound screenings conducted between 11 and 14 weeks of gestation (June 2021 to June 2023). Three experienced radiologists performed fetal screening using 2D video (2D-RAD) and volumes (3D-RAD) to obtain the CRL plane and measurement. The 3DCRL-Net model automatically outputs the landmark position, CRL plane localization and measurement. Three specialists audited the planes achieved by radiologists and 3DCRL-Net as standard or non-standard. The performance of CRL landmark detection, plane localization, and measurement was evaluated in the internal testing dataset, comparing results with 3D-RAD. The performance of CRL plane localization, measurement, and time efficiency were assessed between 3DCRL-Net, 2D-RAD, and 3D-RAD in the external dataset.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe internal dataset consisted of 126 cases in the testing set (training: validation: testing\u0026thinsp;=\u0026thinsp;8:1:1), and the external dataset included 245 cases. On the internal testing dataset, 3DCRL-Net demonstrated a mean absolute distance error of 1.81 mm for the nine landmarks, high accuracy in standard plane localization compared to 3D-RAD (91.27% vs. 80.16%), and strong consistency in CRL measurements, with a 1.26 mm measurement error (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.70). On the external testing dataset, 3DCRL-Net demonstrated high performance in standard plane localization, achieving results comparable to 2D-RAD and 3D-RAD (accuracy: 91.43% vs. 91.43% vs. 86.12%), with strong consistency in CRL measurements, compared to 2D-RAD, which showed a mean absolute error of 1.58 mm and a mean difference of 1.12 mm (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.25).\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe 3DCRL-Net model achieved high performance in CRL evaluation of the standard plane, providing reliable CRL measurements comparable to radiologists using 2D ultrasound and surpassing those using 3D ultrasound, while facilitating CRL evaluation in clinical practice within a few seconds.\u003c/p\u003e","manuscriptTitle":"Comparative Study of 2D vs. 3D AI-Enhanced Ultrasound for Accurate Fetal Crown–rump Length Evaluation in the First Trimester","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-21 08:40:35","doi":"10.21203/rs.3.rs-6250944/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-05-23T13:26:58+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-21T13:00:10+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"220293266172197217750541761310073964780","date":"2025-05-05T07:26:23+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-12T07:54:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"51194977473960439072311626215952295413","date":"2025-04-12T05:04:50+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-03-29T14:03:37+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-03-24T12:04:22+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-03-21T05:07:27+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-03-21T05:04:42+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Pregnancy and Childbirth","date":"2025-03-18T08:14:12+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-pregnancy-and-childbirth","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"prch","sideBox":"Learn more about [BMC Pregnancy and Childbirth](http://bmcpregnancychildbirth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/prch/default.aspx","title":"BMC Pregnancy and Childbirth","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"6277feb0-963a-482b-80ab-ff8010a80e85","owner":[],"postedDate":"April 21st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-07-21T16:01:45+00:00","versionOfRecord":{"articleIdentity":"rs-6250944","link":"https://doi.org/10.1186/s12884-025-07823-6","journal":{"identity":"bmc-pregnancy-and-childbirth","isVorOnly":false,"title":"BMC Pregnancy and Childbirth"},"publishedOn":"2025-07-16 15:57:07","publishedOnDateReadable":"July 16th, 2025"},"versionCreatedAt":"2025-04-21 08:40:35","video":"","vorDoi":"10.1186/s12884-025-07823-6","vorDoiUrl":"https://doi.org/10.1186/s12884-025-07823-6","workflowStages":[]},"version":"v1","identity":"rs-6250944","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6250944","identity":"rs-6250944","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}