Fetal Imaging Research Articles

A benchmark for 2D foetal brain ultrasound analysis

Brain development involves a sequence of structural changes from early stages of the embryo until several months after birth. Currently, ultrasound is the established technique for screening due to its ability to acquire dynamic images in real-time without radiation and to its cost-efficiency. However, identifying abnormalities remains challenging due to the difficulty in interpreting foetal brain images. In this work we present a set of 104 2D foetal brain ultrasound images acquired during the 20th week of gestation that have been co-registered to a common space from a rough skull segmentation. The images are provided both on the original space and template space centred on the ellipses of all the subjects. Furthermore, the images have been annotated to highlight landmark points from structures of interest to analyse brain development. Both the final atlas template with probabilistic maps and the original images can be used to develop new segmentation techniques, test registration approaches for foetal brain ultrasound, extend our work to longitudinal datasets and to detect anomalies in new images.

The New Frontiers of Fetal Imaging: MRI Insights into Cardiovascular and Thoracic Structures.

Fetal magnetic resonance imaging (fMRI) represents a second-line imaging modality that provides multiparametric and multiplanar views that are crucial for confirming diagnoses, detecting associated pathologies, and resolving inconclusive ultrasound findings. The introduction of high-field magnets and new imaging sequences has expanded MRI's role in pregnancy management. Recent innovations in ECG-gating techniques have revolutionized the prenatal evaluation of congenital heart disease by synchronizing imaging with the fetal heartbeat, thus addressing traditional challenges in cardiac imaging. Fetal cardiac MRI (fCMR) is particularly valuable for assessing congenital heart diseases, especially when ultrasound is limited by poor imaging conditions. fCMR allows for detailed anatomical and functional evaluation of the heart and great vessels and is also useful for diagnosing additional anomalies and analyzing blood flow patterns, which can aid in understanding abnormal fetal brain growth and placental perfusion. This review emphasizes fMRI's potential in evaluating cardiac and thoracic structures, including various gating techniques like metric optimized gating, self-gating, and Doppler ultrasound gating. The review also covers the use of static and cine images for structural and functional assessments and discusses advanced techniques like 4D-flow MRI and T1 or T2 mapping for comprehensive flow quantification and tissue characterization.

In Vivo Demonstration of a Real-Time Temporal SNR Acoustic Output Adjustment Method.

This work proposes a novel method of temporal signal-to-noise ratio (SNR)-guided adaptive acoustic output adjustment and demonstrates this approach during in vivo fetal imaging. Acoustic output adjustment is currently the responsibility of sonographers, but ultrasound safety studies show recommended as low as reasonably achievable (ALARA) practices are inconsistently followed. This study explores an automated ALARA method that adjusts the mechanical index (MI) output, targeting imaging conditions matching the temporal noise perception threshold. A 28-dB threshold SNR is used as the target SNR, following prior work showing relevant noise quantities are imperceptible once this image data quality level is reached. After implementing adaptive output adjustment on a clinical system, the average MI required to achieve 28-dB SNR in an 11-volunteer fetal abdomen imaging test ranged from 0.17 to 0.26. The higher MI levels were required when imaging at higher frequencies. During tests with 20-s MI adjustment imaging periods, the degree of motion impacted the adaptive performance. For stationary imaging views, target SNR levels were maintained in 90% of SNR evaluations. When scanning between targets the imaging conditions were more variable, but the target SNR was still maintained in 71% of the evaluations. Given the relatively low MI recommended when performing MI adjustment and the successful adjustment of MI in response to changing imaging conditions, these results encourage adoption of adaptive acoustic output approaches guided by temporal SNR.

Impact of maternal diabetes mellitus on fetal atrial strain.

While Maternal Diabetes Mellitus (DM) is well known to affect the size and function of multiple fetal organ systems, effects on developing heart chamber function remain difficult to assess. We sought to determine the independent impact of maternal DM on fetal cardiac function in middle pregnancy. We prospectively recruited mothers with all categories of DM and non-diabetic healthy controls (NDC). Echocardiograms were optimized for chamber quantification and strain analysis. Left atrial area (LAA), LA strain (LAS), right atrial strain (RAS), global longitudinal ventricular strain (GLS) and Right ventricular free wall strain (RV FWS) were evaluated by 2 blinded operators. After excluding 9 mothers with poor fetal image quality, images from 104 mothers with DM and 47 NDC were analyzed. Mothers with DM and NDCs were well matched for age, blood pressure, smoking prevalence, and gestational age. Fetal heart rate (FHR) was significantly higher in fetuses of mothers with DM compared to NDC (147 ± 10bpm vs. 144 ± 8, p = 0.04). LAA in fetuses of mothers with DM trended towards being larger in size (1.68 ± 0.4cm2 vs. 1.56 ± 0.4cm2, p = 0.08). Fetal septal diameters were larger in maternal DM compared to NDC (2.7 ± 0.5cm vs. 2.5 ± 0.5cm, p = 0.001). GLS was similar between the groups. Fetal LAS was lower in maternal DM (28.8 ± 8.8% vs. 33.3 ± 10.4%, p = 0.007) and was independently associated with maternal DM after adjusting for GLS and FHR. Fetal RAS was lower in maternal DM (27.7 ± 10.4% vs. 31.8 ± 10.3%, p = 0.007), however only determinates were estimated fetal weight and RV FWS. Maternal DM independently impairs fetal LA function in mid pregnancy. These early functional changes in the developing heart warrant future studies investigating impact on cardiovascular health.

FetMRQC: A robust quality control system for multi-centric fetal brain MRI

Fetal brain MRI is becoming an increasingly relevant complement to neurosonography for perinatal diagnosis, allowing fundamental insights into fetal brain development throughout gestation. However, uncontrolled fetal motion and heterogeneity in acquisition protocols lead to data of variable quality, potentially biasing the outcome of subsequent studies. We present FetMRQC, an open-source machine-learning framework for automated image quality assessment and quality control that is robust to domain shifts induced by the heterogeneity of clinical data. FetMRQC extracts an ensemble of quality metrics from unprocessed anatomical MRI and combines them to predict experts’ ratings using random forests. We validate our framework on a pioneeringly large and diverse dataset of more than 1600 manually rated fetal brain T2-weighted images from four clinical centers and 13 different scanners. Our study shows that FetMRQC’s predictions generalize well to unseen data while being interpretable. FetMRQC is a step towards more robust fetal brain neuroimaging, which has the potential to shed new insights on the developing human brain.

Efficient segmentation of fetal brain MRI based on the physical resolution.

The image resolution of fetal brain magnetic resonance imaging (MRI) is a critical factor in brain development measures, which is mainly determined by the physical resolution configured in the MRI sequence. However, fetal brain MRI are commonly reconstructed to 3D images with a higher apparent resolution, compared to the original physical resolution. This work is to demonstrate that accurate segmentation can be achieved based on the MRI physical resolution, and the high apparent resolution segmentation can be achieved by a simple deep learning module. This retrospective study included 150 adult and 80 fetal brain MRIs. The adult brain MRIs were acquired at a high physical resolution, which were downsampled to visualize and quantify its impacts on the segmentation accuracy. The physical resolution of fetal images was estimated based on MRI acquisition settings and the images were downsampled accordingly before segmentation and restored using multiple upsampling strategies. Segmentation accuracy of ConvNet models were evaluated on the original and downsampled images. Dice coefficients were calculated, and compared to the original data. When the apparent resolution was higher than the physical resolution, the accuracy of fetal brain segmentation had negligible degradation (accuracy reduced by 0.26%, 1.1%, and 1.8% with downsampling factors of 4/3, 2, and 4 in each dimension, without significant differences from the original data). Using a downsampling factor of 4 in each dimension, the proposed method provided 7× smaller and 10× faster models. Efficient and accurate fetal brain segmentation models can be developed based on the physical resolution of MRI acquisitions.

Deep Learning Techniques for a Comprehensive Analysis of Fetal Biometric Parameters Across Trimesters

The process of creating fetal images from the uterus using sound influence is known as fetal ultrasound imaging. During this scan, measurements such as the gestational sac, biparietal diameter, head circumference, abdominal circumference, and femur length can be taken from the mother, which are further analyzed by the radiologist or gynecologist. These factors allow us to quickly test for anomalies and monitor the fetal growth and development of a baby. This paper delves into the techniques utilized in previous studies for analyzing abnormalities from ultrasound images using machine learning and deep learning techniques. Specifically, we focus on two trimesters and three key fetal parameters: Head Circumference (HC), Abdominal Circumference (AC), and Femur Length (FL). Our proposed method, the Unet segmentation method, not only performs segmentation but also predicts the parameters. We employ various transfer learning techniques for classification. The experiment involves 1,313 medical fetal images, comprising 563 from the second trimester and 750 from the third trimester. In terms of segmentation accuracy, the results for the second trimester's biometric parameters are as follows: AC = 69.09%, FL = 92.02%, and HC = 69.43%. For the third trimester, the accuracy for FL is 90.04%, and for HC, it is 69.76%. Regarding classification methods, MobileNet and XceptionNet yield comparable results. For the second trimester, MobileNet achieves 99.28%, and XceptionNet achieves 99.82%. For the third trimester, both MobileNet and XceptionNet achieve 99.86%.

Automatic Detection of Standard Planes in Fetal Ultrasound Images based on Convolutional Neural Networks and Ensemble Learning

aims: This study aims to leverage artificial intelligence for enhancing medical diagnosis, focusing on ultrasound evaluation of fetal development and detection of fetal diseases. background: Traditional diagnostic methods in ultrasound are known for being time-consuming and laborious, prompting the need for more efficient approaches. objective: The objective of this research is to develop an end-to-end automatic diagnosis system using convolutional neural networks with ensemble learning to enhance robustness and accuracy in classifying ultrasound images. method: The study involves constructing and implementing the automatic diagnosis system, training it on a diverse dataset encompassing six categories: abdomen, brain, femur, thorax, maternal cervix, and other planes. result: Experimental results demonstrate that the proposed end-to-end system significantly improves the detection accuracy of the standard plane in ultrasound images. conclusion: The application of artificial intelligence through an ensemble learning-based automatic diagnosis system shows promise in advancing ultrasound-based medical diagnosis, particularly in fetal development assessment. other: This research contributes to the ongoing efforts in leveraging technology for more efficient and accurate medical diagnostic processes.

Ensemble learning for fetal ultrasound and maternal–fetal data to predict mode of delivery after labor induction

Providing adequate counseling on mode of delivery after induction of labor (IOL) is of utmost importance. Various AI algorithms have been developed for this purpose, but rely on maternal–fetal data, not including ultrasound (US) imaging. We used retrospectively collected clinical data from 808 subjects submitted to IOL, totaling 2024 US images, to train AI models to predict vaginal delivery (VD) and cesarean section (CS) outcomes after IOL. The best overall model used only clinical data (F1-score: 0.736; positive predictive value (PPV): 0.734). The imaging models employed fetal head, abdomen and femur US images, showing limited discriminative results. The best model used femur images (F1-score: 0.594; PPV: 0.580). Consequently, we constructed ensemble models to test whether US imaging could enhance the clinical data model. The best ensemble model included clinical data and US femur images (F1-score: 0.689; PPV: 0.693), presenting a false positive and false negative interesting trade-off. The model accurately predicted CS on 4 additional cases, despite misclassifying 20 additional VD, resulting in a 6.0% decrease in average accuracy compared to the clinical data model. Hence, integrating US imaging into the latter model can be a new development in assisting mode of delivery counseling.

HCN: Hybrid Capsule Network for Fetal Plane Classification in Ultrasound Images

ABSTRACTClassifying fetal ultrasound images into different anatomical categories, such as the abdomen, brain, femur, thorax, and so forth can contribute to the early identification of potential anomalies or dangers during prenatal care. Ignoring major abnormalities that might lead to fetal death or permanent disability. This article proposes a novel hybrid capsule network architecture‐based method for identifying fetal ultrasound images. The proposed architecture increases the precision of fetal image categorization by combining the benefits of a capsule network with a convolutional neural network. The proposed hybrid model surpasses conventional convolutional network‐based techniques with an overall accuracy of 0.989 when tested on a publicly accessible dataset of prenatal ultrasound images. The results indicate that the proposed hybrid architecture is a promising approach for precisely and consistently classifying fetal ultrasound images, with potential uses in clinical settings.

A Coarse-Fine Collaborative Learning Model for Three Vessel Segmentation in Fetal Cardiac Ultrasound Images.

Congenital heart disease (CHD) is the most frequent birth defect and a leading cause of infant mortality, emphasizing the crucial need for its early diagnosis. Ultrasound is the primary imaging modality for prenatal CHD screening. As a complement to the four-chamber view, the three-vessel view (3VV) plays a vital role in detecting anomalies in the great vessels. However, the interpretation of fetal cardiac ultrasound images is subjective and relies heavily on operator experience, leading to variability in CHD detection rates, particularly in resource-constrained regions. In this study, we propose an automated method for segmenting the pulmonary artery, ascending aorta, and superior vena cava in the 3VV using a novel deep learning network named CoFi-Net. Our network incorporates a coarse-fine collaborative strategy with two parallel branches dedicated to simultaneous global localization and fine segmentation of the vessels. The coarse branch employs a partial decoder to leverage high-level semantic features, enabling global localization of objects and suppression of irrelevant structures. The fine branch utilizes attention-parameterized skip connections to improve feature representations and improve boundary information. The outputs of the two branches are fused to generate accurate vessel segmentations. Extensive experiments conducted on a collected dataset demonstrate the superiority of CoFi-Net compared to state-of-the-art segmentation models for 3VV segmentation, indicating its great potential for enhancing CHD diagnostic efficiency in clinical practice. Furthermore, CoFi-Net outperforms other deep learning models in breast lesion segmentation on a public breast ultrasound dataset, despite not being specifically designed for this task, demonstrating its potential and robustness for various segmentation tasks.

Optimizing Performance of Transformer-based Models for Fetal Brain MR Image Segmentation.

Purpose To test the performance of a transformer-based model when manipulating pretraining weights, dataset size, and input size and comparing the best model with the reference standard and state-of-the-art models for a resting-state functional (rs-fMRI) fetal brain extraction task. Materials and Methods An internal retrospective dataset (172 fetuses, 519 images; collected 2018-2022) was used to investigate influence of dataset size, pretraining approaches, and image input size on Swin-U-Net transformer (UNETR) and UNETR models. The internal and external (131 fetuses, 561 images) datasets were used to cross-validate and to assess generalization capability of the best model versus state-of-the-art models on different scanner types and number of gestational weeks (GWs). The Dice similarity coefficient (DSC) and the balanced average Hausdorff distance (BAHD) were used as segmentation performance metrics. Generalized equation estimation multifactorial models were used to assess significant model and interaction effects of interest. Results The Swin-UNETR model was not affected by the pretraining approach and dataset size and performed best with the mean dataset image size, with a mean DSC of 0.92 and BAHD of 0.097. Swin-UNETR was not affected by scanner type. Generalization results on the internal dataset showed that Swin-UNETR had lower performance compared with the reference standard models and comparable performance on the external dataset. Cross-validation on internal and external test sets demonstrated better and comparable performance of Swin-UNETR versus convolutional neural network architectures during the late-fetal period (GWs > 25) but lower performance during the midfetal period (GWs ≤ 25). Conclusion Swin-UNTER showed flexibility in dealing with smaller datasets, regardless of pretraining approaches. For fetal brain extraction from rs-fMR images, Swin-UNTER showed comparable performance with that of reference standard models during the late-fetal period and lower performance during the early GW period. Keywords: Transformers, CNN, Medical Imaging Segmentation, MRI, Dataset Size, Input Size, Transfer Learning Supplemental material is available for this article. © RSNA, 2024.

Introducing and Validating the Cranial-Dorsal-Hip Angle (∠CDH): A Method for Accurate Fetal Position Assessment in the First Trimester and Future AI Applications

Abstract Purpose To introduce the cranial-dorsal-hip angle (∠CDH) as a novel quantitative tool for assessing fetal position in the first trimester and to validate its feasibility for future AI applications. Materials and Methods 2520 first-trimester fetal NT exams with 2582 CRL images (January-August 2022) were analyzed at a tertiary hospital as the pilot group. Additionally, 1418 cases with 1450 fetal CRL images (September-December 2022) were examined for validation. Three expert sonographers defined a standard for fetal positions. ∠CDH measurements, conducted by two ultrasound technicians, were validated for consistency using Bland-Altman plots and the intra-class correlation coefficient (ICC). This method allowed for categorizing fetal positions as hyperflexion, neutral, and hyperextension based on ∠CDH. Comparative accuracy was assessed against Ioannou, Wanyonyi, and Roux methods using the weighted Kappa coefficient (k value). Results The pilot group comprised 2186 fetal CRL images, and the validation group included 1193 images. Measurement consistency was high (ICCs of 0.993; P<0.001). The established 95% reference range for ∠CDH in the neutral fetal position was 118.3° to 137.8°. The ∠CDH method demonstrated superior accuracy over the Ioannou, Wanyonyi, and Roux methods in both groups, with accuracy rates of 94.5% (k values: 0.874, 95%CI: 0.852–0.896) in the pilot group, and 92.6% (k values: 0.838, 95%CI: 0.806–0.871) in the validation group. Conclusion The ∠CDH method has been validated as a highly reproducible and accurate technique for first-trimester fetal position assessment. This sets the stage for its potential future integration into intelligent assessment models.

Proposal for standardized prenatal assessment of fetal open dysraphisms by the European reference network for Intellectual disability, TeleHealth, Autism and Congenital Anomalies (ITHACA) and eUROGEN.

Open dysraphisms, that is, myelomeningocele and myeloschisis, are rare diseases associated with a risk of severe disability, including lower limb motor and sensory deficiency, sphincter deficiency, and potential intellectual deficiency. Open dysraphism is diagnosed in Europe in 93.5% of cases. In case of suspicion of fetal open dysraphism, a detailed fetal morphologic assessment is required to confirm the diagnosis and exclude associated structural anomalies, as well as genetic assessment. In case of isolated fetal open dysraphism, assessment of prognosis is based on fetal imaging including the level of the lesion, the presence or not of a sac, the presence and nature of intra cranial anomalies, and the anatomical and functional evaluation of the lower extremities. Based on these biomarkers, a personalized prognosis as well as comprehensive information about prenatal management alternatives will allow parents to decide on further management options. Standardization of prenatal assessment is essential to compare outcomes with benchmark data and make assessment of surgical innovation possible. Herein, we propose a protocol for the standardized ultrasound assessment of fetuses with isolated open dysraphism.

Relationship between the sonographic foetal lung-to-liver ratio of mean grey level and progesterone concentration in canine late gestation

A study was conducted to perform the quantitative analysis of canine foetal lung and liver sonographic images in the late term of gestation (58-67 days) and to evaluate its association with serum progesterone concentration. Thirty-two pregnant bitches were selected for the study. Lung development of foetus occurs in three different stages viz the pseudo glandular (40-48), canalicular (49-56 d) and saccular phase (57-63 d) of pregnancy. Lungs of the foetus at the time of parturition are in saccular phase of development and the alveolar phase occurs in the neonatal period. In the present study, sonographic changes occurring in the lung during the saccular phase of foetal lung development was evaluated. Mean grey level of the foetal lung and liver sonographic images were estimated using dedicated image analysis software Image J. The lung-to-liver ratio (LLR) of mean grey level (MGL) was recorded. Corresponding progesterone concentrations were also estimated for each animal. Correlation statistical analysis revealed a significant correlation between LLR of MGL and progesterone concentration in the animals selected for the study. LLR of MGL is a prognostic indicator of foetal survival and since LLR of MGL and progesterone are positively correlated, it can be used as an indicator of foetal maturity in place of progesterone concentration. Keywords: Lung-to-liver ratio of mean grey level, ultrasonography, progesterone

Automatic segmentation based on optimization U-Net neural network (OU-NetNN) for fetal cardiac ultrasound images

Automatic segmentation based on optimization U-Net neural network (OU-NetNN) for fetal cardiac ultrasound images

Imaging Assessment of Fetal Cerebrovascular Anomalies.

Four distinct vascular anomalies can be seen to affect the brain on fetal imaging: vein of Galen malformations, non-galenic arteriovenous pial fistulas, dural sinus malformations, and intracranial venous malformations. These congenital disorders affect the arteries and veins of the developing brain and are rarely seen beyond the neonatal stage. The four fetal cerebrovascular anomalies are associated with quite disparate natural histories and prognoses. MRI plays a pivotal role in the evaluation of fetuses with these conditions due to its ability to definitively establish the diagnosis, to detect subtle parenchymal injuries, to delineate the course of abnormal vessels in detail and to some extent the nature of vascular flow, and to identify ischemic, thrombotic, and hemorrhagic complications. Recently, an investigational transurterine embolization procedure targeted at treating fetuses with vein of Galen malformations who are at high risk for neonatal decompensation has emerged as a promising alternative to expectant management and post-natal embolization, with imaging being used to identify suitable patients for the intervention and in pre-procedural planning. This manuscript reviews the essential imaging and clinical features of these four fetal neurovascular anomalies and underscores the practical aspects related to counseling, prognosis, and the multidisciplinary management of these entities.ABBREVIATIONS: ACVRL1= activin A receptor like type 1; b-SSFP=Balanced Steady State Free precession; DSM= Dural Sinus Malformation; Ephrin B4= Ephrin type-B receptor 4; icVM= Intracranial Venous Malformation; ITGB1= Integrin Subunit Beta 1; NOTCH1= Neurogenic locus notch homolog protein 1; PTPN11= Protein Tyrosine Phosphatase Non-Receptor Type 11; RASA1= RAS P21 Protein Activator 1; SSFSE= Single-shot fast spin echo; VOGM=Vein of Galen Malformation.

The Fetal Spleen in Low-Risk Pregnancies and prior to Preterm Birth: Observational Study of the Role of Anatomical and Functional Magnetic Resonance Imaging

Introduction: Spontaneous preterm birth complicates ∼7% of pregnancies and causes morbidity and mortality. Although infection is a common etiology, our understanding of the fetal immune system in vivo is limited. This study aimed to utilize T2-weighted imaging and T2* relaxometry (which is a proxy of tissue oxygenation) of the fetal spleen in uncomplicated pregnancies and in fetuses that were subsequently delivered spontaneously prior to 32 weeks. Methods: Women underwent imaging including T2-weighted fetal body images and multi-eco gradient echo single-shot echo planar sequences on a Phillips Achieva 3T system. Previously described postprocessing techniques were applied to obtain T2- and T2*-weighted imaging of the fetal spleen and T2-weighted fetal body volumes. Results: Among 55 women with uncomplicated pregnancies, an increase in fetal splenic volume, splenic:body volume, and a decrease in splenic T2* signal intensity was demonstrated across gestation. Compared to controls, fetuses who were subsequently delivered prior to 32 weeks’ gestation (n = 19) had a larger spleen when controlled for the overall size of the fetus (p = 0.027), but T2* was consistent (p = 0.76). Conclusion: These findings provide evidence of a replicable method of studying the fetal immune system and give novel results on the impact of impending preterm birth on the spleen. While T2* decreases prior to preterm birth in other organs, preservation demonstrated here suggests preferential sparing of the spleen.

Associations and outcomes of prenatally detected rhombencephalosynapsis.

To describe the association between prenatal imaging and neurodevelopmental outcomes of fetuses with rhombencephalosynapsis (RES). Thirty-four pregnancies complicated by RES were identified from our institutional databases based on US and/or MRI findings. Genetic testing results were gathered. In cases of termination of pregnancy, we studied the association between prenatal imaging and neuropathologic findings. For those who opted for expectant management, comprehensive developmental assessments and postnatal MRI imaging were evaluated. Over one third of fetuses in our cohort had complete RES. Common intracranial anomalies identified were mesencephalosynapsis, aqueduct stenosis and diencephalosynapsis. The degree of RES was not associated with the frequency of additional central nervous system anomalies. MRI had a good correlation with neuropathologic findings with regard to the degree of RES, aqueduct stenosis and mesencephalosynapsis. Postmortem autopsy showed that one third of our cases had VACTERL-H and almost all of those had complete RES. All liveborn neonates(n=6) had aqueduct stenosis requiring ventriculoperitoneal shunting within days of delivery (median 5days). While a large proportion of prenatally suspected complete RES were found to have partial RES on postnatal imaging, prenatal diagnosis of aqueduct stenosis remained unchanged. All children that were at least 2years old (n=3) had global developmental delay. Prenatal assessment of the RES severity is challenging and may be unreliable. Nevertheless, postnatal prognosis is poor for both complete and partial RES. Associated aqueductal stenosis, can be reliably assessed prenatally and this may contribute to worse postnatal prognosis than the degree of RES.

Deep learning-based differentiation of ventricular septal defect from tetralogy of Fallot in fetal echocardiography images.

Congenital heart disease (CHD) seriously affects children's health and quality of life, and early detection of CHD can reduce its impact on children's health. Tetralogy of Fallot (TOF) and ventricular septal defect (VSD) are two types of CHD that have similarities in echocardiography. However, TOF has worse diagnosis and higher morality than VSD. Accurate differentiation between VSD and TOF is highly important for administrative property treatment and improving affected factors' diagnoses. TOF and VSD were differentiated using convolutional neural network (CNN) models that classified fetal echocardiography images. We collected 105 fetal echocardiography images of TOF and 96 images of VSD. Four CNN models, namely, VGG19, ResNet50, NTS-Net, and the weakly supervised data augmentation network (WSDAN), were used to differentiate the two congenital heart diseases. The performance of these four models was compared based on sensitivity, accuracy, specificity, and AUC. VGG19 and ResNet50 performed similarly, with AUCs of 0.799 and 0.802, respectively. A superior performance was observed with NTS-Net and WSDAN specific for fine-grained image categorization tasks, with AUCs of 0.823 and 0.873, respectively. WSDAN had the best performance among all models tested. WSDAN exhibited the best performance in differentiating between TOF and VSD and is worthy of further clinical popularization.