Interaction between clinicians and artificial intelligence to detect fetal atrioventricular septal defects on ultrasound: how can we optimize collaborative performance?

OBJECTIVES: Artificial intelligence (AI) has shown promise in improving the performance of fetal ultrasound screening in detecting congenital heart disease (CHD). The effect of giving AI advice to human operators has not been studied in this context. Giving additional information about AI model workings, such as confidence scores for AI predictions, may be a way of further improving performance. Our aims were to investigate whether AI advice improved overall diagnostic accuracy (using a single CHD lesion as an exemplar), and to determine what, if any, additional information given to clinicians optimized the overall performance of the clinician-AI team. METHODS: An AI model was trained to classify a single fetal CHD lesion (atrioventricular septal defect (AVSD)), using a retrospective cohort of 121 130 cardiac four-chamber images extracted from 173 ultrasound scan videos (98 with normal hearts, 75 with AVSD); a ResNet50 model architecture was used. Temperature scaling of model prediction probability was performed on a validation set, and gradient-weighted class activation maps (grad-CAMs) produced. Ten clinicians (two consultant fetal cardiologists, three trainees in pediatric cardiology and five fetal cardiac sonographers) were recruited from a center of fetal cardiology to participate. Each participant was shown 2000 fetal four-chamber images in a random order (1000 normal and 1000 AVSD). The dataset comprised 500 images, each shown in four conditions: (1) image alone without AI output; (2) image with binary AI classification; (3) image with AI model confidence; and (4) image with grad-CAM image overlays. The clinicians were asked to classify each image as normal or AVSD. RESULTS: A total of 20 000 image classifications were recorded from 10 clinicians. The AI model alone achieved an accuracy of 0.798 (95% CI, 0.760-0.832), a sensitivity of 0.868 (95% CI, 0.834-0.902) and a specificity of 0.728 (95% CI, 0.702-0.754), and the clinicians without AI achieved an accuracy of 0.844 (95% CI, 0.834-0.854), a sensitivity of 0.827 (95% CI, 0.795-0.858) and a specificity of 0.861 (95% CI, 0.828-0.895). Showing a binary (normal or AVSD) AI model output resulted in significant improvement in accuracy to 0.865 (P < 0.001). This effect was seen in both experienced and less-experienced participants. Giving incorrect AI advice resulted in a significant deterioration in overall accuracy, from 0.761 to 0.693 (P < 0.001), which was driven by an increase in both Type-I and Type-II errors by the clinicians. This effect was worsened by showing model confidence (accuracy, 0.649; P < 0.001) or grad-CAM (accuracy, 0.644; P < 0.001). CONCLUSIONS: AI has the potential to improve performance when used in collaboration with clinicians, even if the model performance does not reach expert level. Giving additional information about model workings such as model confidence and class activation map image overlays did not improve overall performance, and actually worsened performance for images for which the AI model was incorrect. © 2024 The Authors. Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of International Society of Ultrasound in Obstetrics and Gynecology.

Measurement of the cardiac time intervals of the fetal ECG utilising a computerised algorithm: A retrospective observational study.

Objective: Establish whether the reliable measurement of cardiac time intervals of the fetal ECG can be automated and to address whether this approach could be used to investigate large datasets. Design: Retrospective observational study. Setting: Teaching hospitals in London UK, Nottingham UK and New York USA. Participants: Singleton pregnancies with no known fetal abnormality. Methods: Archived fetal ECG's performed using the MonicaAN24 monitor. A single ECG (PQRST) complex was generated from 5000 signal-averaged beats and electrical cardiac time intervals measured in an automated way and manually. Main Outcome measure: Validation of a newly developed algorithm to measure the cardiac time intervals of the fetal ECG. Results: 188/236 (79.7%) subjects with fECGs of suitable signal:noise ratio were included for analysis comparing manual with automated measurement. PR interval was measured in 173/188 (92%), QRS complex in 170/188 (90%) and QT interval in 123/188 (65.4%). PR interval was 107.6 (12.07) ms [mean(SD)] manual vs 109.11 (14.7) ms algorithm. QRS duration was 54.72(6.35) ms manual vs 58.34(5.73) ms algorithm. QT-interval was 268.93 (21.59) ms manual vs 261.63 (36.16) ms algorithm. QTc was 407.5(32.71) ms manual vs 396.4 (54.78) ms algorithm. The QRS-duration increased with gestational age in both manual and algorithm measurements. Conclusion: Accurate measurement of fetal ECG cardiac time intervals can be automated with potential application to interpretation of larger datasets.

3D/4D spatiotemporal image correlation (STIC) fetal echocardiography provides incremental benefit over 2D fetal echocardiography in predicting postnatal surgical approach in double-outlet right ventricle.

OBJECTIVE: To analyze the incremental benefit of 3D/4D spatiotemporal image correlation (STIC) fetal echocardiography over 2D fetal echocardiography with respect to the accuracy of identification of anatomic details crucial for surgical decision-making and in predicting surgical approach in fetuses with double-outlet right ventricle (DORV). METHODS: This was a retrospective study of fetuses with DORV which had undergone both 2D echocardiography and 3D/4D STIC echocardiography and which underwent surgery postnatally in a tertiary pediatric cardiac center in Kerala between October 2015 and March 2019. All such cases with normal atrial arrangement, concordant atrioventricular connections and balanced ventricles were included. 2D and 3D/4D STIC fetal echocardiographic data were analyzed by two experienced observers blinded to the other dataset. Anatomic variables crucial for surgical decision-making, i.e. location and routability of the ventricular septal defect, relationship of the great arteries and presence of outflow obstruction, were compared between the two modalities with respect to agreement with postnatal echocardiography. The accuracy of prenatal prediction of the surgical pathway was compared between 2D and 3D/4D modalities with respect to the procedure undertaken. RESULTS: Included in the study were 22 fetuses with DORV which had undergone both 2D and 3D/4D imaging as well as postnatal surgery. Accuracy of prenatal interpretation of all four anatomic variables was significantly higher using 3D/4D STIC than using 2D fetal echocardiography (19/22 (86.4%) vs 8/22 (36.4%), P < 0.001). Surgical procedures included single-stage repair in 14 (63.5%) patients and a multistage approach in eight (36.4%). Prenatal prediction of the surgical pathway was significantly more accurate on 3D/4D STIC than on 2D echocardiography (20/22 (90.9%) vs 12/22 (54.5%), P = 0.021). Prenatal predictive accuracy of single-stage biventricular repair was significantly better for 3D/4D STIC than for 2D echocardiography (14/14 (100%) vs 8/14 (57.1%), P = 0.04). CONCLUSION: Addition of 3D/4D STIC to conventional 2D fetal echocardiography confers incremental benefit on the accuracy of identification of anatomic details crucial for surgical decision-making and the prediction of postnatal surgical approach in fetuses with DORV, thereby potentially aiding prenatal counseling. © 2020 International Society of Ultrasound in Obstetrics and Gynecology.

Impact of prospective measurement of outflow tracts in prediction of coarctation of the aorta.

OBJECTIVES: Prenatal diagnosis of coarctation of the aorta (CoA) is associated with reduced mortality and morbidity, however, accurate prenatal prediction remains challenging. To date, studies have used retrospective measurements of the outflow tracts to evaluate their potential to predict CoA. Our primary objective was to evaluate prospectively acquired measurements of the outflow tracts in fetuses with prenatally suspected CoA. A secondary aim was to report the postnatal prevalence of bicuspid aortic valve in this cohort. METHODS: Pregnancies with suspicion of isolated CoA and with a minimum of 6 months' postnatal follow-up available were identified from the cardiac database of a tertiary fetal cardiology center in the UK, between January 2002 and December 2017. Measurement of the aortic valve, pulmonary valve, distal transverse aortic arch (DTAA) and arterial duct (AD) diameters were undertaken routinely in fetuses with suspected CoA during the study period. Z-scores were computed using published reference ranges based on > 7000 fetuses from our own unit. RESULTS: Of 149 pregnancies with prenatally suspected CoA included in the study, CoA was confirmed within 6 months after birth in 77/149 (51.7%) cases. DTAA diameter Z-score and the Z-score of second-trimester DTAA/AD diameter ratio were smaller in fetuses with postnatally confirmed CoA than those in false-positive cases (-2.8 vs -1.9; P = 0.039 and -3.13 vs -2.61; P = 0.005, respectively). Multiple regression analysis demonstrated that the Z-scores of DTAA and AD diameters were the only significant predictors of postnatal CoA (P = 0.001). Bicuspid aortic valve was identified in 30% of the false-positive cases. CONCLUSIONS: Measurement of DTAA and AD diameter Z-scores can be used to ascertain risk for postnatal CoA in a selected cohort. The high incidence of bicuspid aortic valve in false-positive cases merits further study with respect to both etiology and longer-term significance. Copyright © 2019 ISUOG. Published by John Wiley & Sons Ltd.

Myocardial deformation in fetuses with coarctation of the aorta: a case-control study.

OBJECTIVE: To study myocardial deformation by speckle tracking echocardiography in fetuses with coarctation of the aorta (CoA) compared with gestational age-matched normal controls. METHODS: This was a retrospective study of 12 fetuses with postnatally confirmed CoA and 12 gestational age-matched controls seen at a tertiary fetal cardiology unit between January 2013 and July 2014. Two-dimensional speckle tracking in standard grayscale four-chamber view of the fetal heart was performed to assess left and right myocardial deformation. Global longitudinal strain, strain rate and velocities, and regional longitudinal strain were analyzed and compared with controls. RESULTS: Median gestational age at echocardiography was 25 + 4 weeks. Fetuses with CoA presented with a narrower, but not shorter, left ventricle when compared with controls (mitral valve diastolic diameter, 5.90 vs 8.50 mm; P = 0.002; left ventricular diastolic length, 16.50 vs 18.50 mm; P = 0.05). Global longitudinal systolic strain (P = 0.004), systolic strain rate (P = 0.01) and diastolic strain rate (P = 0.004) of the left ventricle were significantly lower in fetuses with CoA compared with controls. Similar findings were observed for longitudinal systolic (P = 0.03) and diastolic (P = 0.01) velocities of the left ventricle. Right ventricular parameters were not different between groups. CONCLUSIONS: Fetuses with CoA have lower left ventricular longitudinal systolic strain, systolic strain rate and diastolic strain rate when compared with gestational age-matched control fetuses. These differences in deformation might explain, at least in part, the cardiac asymmetry observed in fetuses with CoA. Copyright © 2016 ISUOG. Published by John Wiley & Sons Ltd.

A novel methodology for personalized simulations of ventricular hemodynamics from noninvasive imaging data.

Current state-of-the-art imaging techniques can provide quantitative information to characterize ventricular function within the limits of the spatiotemporal resolution achievable in a realistic acquisition time. These imaging data can be used to personalize computer models, which in turn can help treatment planning by quantifying biomarkers that cannot be directly imaged, such as flow energy, shear stress and pressure gradients. To date, computer models have typically relied on invasive pressure measurements to be made patient-specific. When these data are not available, the scope and validity of the models are limited. To address this problem, we propose a new methodology for modeling patient-specific hemodynamics based exclusively on noninvasive velocity and anatomical data from 3D+t echocardiography or Magnetic Resonance Imaging (MRI). Numerical simulations of the cardiac cycle are driven by the image-derived velocities prescribed at the model boundaries using a penalty method that recovers a physical solution by minimizing the energy imparted to the system. This numerical approach circumvents the mathematical challenges due to the poor conditioning that arises from the imposition of boundary conditions on velocity only. We demonstrate that through this technique we are able to reconstruct given flow fields using Dirichlet only conditions. We also perform a sensitivity analysis to investigate the accuracy of this approach for different images with varying spatiotemporal resolution. Finally, we examine the influence of noise on the computed result, showing robustness to unbiased noise with an average error in the simulated velocity approximately 7% for a typical voxel size of 2mm(3) and temporal resolution of 30ms. The methodology is eventually applied to a patient case to highlight the potential for a direct clinical translation.

Prenatal diagnosis of left ventricular diverticulum and coarctation of the aorta.

Congenital left ventricular diverticulum (LVD) is a rare abnormality of the myocardium which has been detected previously in the fetus. Lesions have been reported from as early as 12 weeks' gestation but are more commonly detected in the mid-second trimester. Fetal presentation of LVD ranges from an abnormal four-chamber view of the heart, arrhythmia or isolated pericardial effusion to fetal hydrops with associated heart failure. Here, we describe the prenatal diagnosis of an infant with LVD originating from the left ventricular outflow tract associated with coarctation of the aorta. The diagnosis was confirmed postnatally by two-dimensional echocardiography and cardiac magnetic resonance imaging. We hypothesize that the lesion compromised antegrade flow into the transverse aortic arch, which may have contributed to underdevelopment of the aortic arch and subsequently the development of coarctation of the aorta. This is a unique case of LVD and coarctation of the aorta.

Accuracy of the transpulmonary ultrasound dilution method for detection of small anatomic shunts.

The purpose of this study was to investigate the qualitative and quantitative accuracy of transpulmonary ultrasound dilution (UD) (COstatus™, Transonic Systems) for the detection of small anatomic shunts. It was a prospective, observational study in a multi-disciplinary pediatric intensive care unit. Seventy-three critically ill children (67 post cardiac surgery), with a median (IQR) age of 10 (3-50.3) months and a median (IQR) weight of 8 (3.43-13) kg were enrolled. Ultrasound dilution (UD) measurements were performed on patients within 1 h of undergoing two-dimensional echocardiography, which was used as the comparator technique. Shunt was diagnosed by characteristic changes on the UD curve shape, and was considered "test-positive" only if two or more measurements suggested the presence of the shunt. The UD technology also provided an estimate of pulmonary to systemic blood flow ratio (Qp:Qs). 12/73 (16.4 %) patients had a shunt identified by both UD and echocardiography. The overall accuracy (95 % CI) was 86.1 % (75.6-96.6 %), with a sensitivity of 85.7 % (57.2-98.2 %) and specificity of 86.4 % (75.0-94.0 %). The estimated Qp:Qs ranged from 0.7 to 1.4, which was consistent qualitatively with the echocardiographic findings on color flow doppler. Shunt was detected by UD alone in eight children; six of these had clinical conditions known to compromise dilution curve analysis (valve regurgitation, asymmetric pulmonary blood flow). Shunt was detected by echocardiography alone in two children; in both cases the shunt was tiny. UD is an accurate method for the detection of small anatomical shunts, both qualitatively and quantitatively.

Towards a fast and efficient approach for modelling the patient-specific ventricular haemodynamics.

Computer modelling of the heart has emerged over the past decade as a powerful technique to explore the cardiovascular pathophysiology and inform clinical diagnosis. The current state-of-the-art in biophysical modelling requires a wealth of, potentially invasive, clinical data for the parametrisation and validation of the models, a process that is still too long and complex to be compatible with the clinical decision-making time. Therefore, there remains a need for models that can be quickly customised to reconstruct physical processes difficult to measure directly in patients. In this paper, we propose a less resource-intensive approach to modelling, whereby computational fluid-dynamics (CFD) models are constrained exclusively by boundary motion derived from imaging data through a validated wall tracking algorithm. These models are generated and parametrised based solely on ultrasound data, whose acquisition is fast, inexpensive and routine in all patients. To maximise the time and computational efficiency, a semi-automated pipeline is embedded in an image processing workflow to personalise the models. Applying this approach to two patient cases, we demonstrate this tool can be directly used in the clinic to interpret and complement the available clinical data by providing a quantitative indication of clinical markers that cannot be easily derived from imaging, such as pressure gradients and the flow energy.

Outcome and requirement for surgical repair following prenatal diagnosis of ventricular septal defect.

OBJECTIVE: To document outcome following prenatal diagnosis of ventricular septal defects (VSDs), particularly associated anomalies and the requirement for surgical closure of the defect. METHODS: All cases of prenatal diagnosis of a VSD made by fetal cardiologists at a tertiary fetal medicine referral center in the period January 2002 to December 2011 were extracted from our database. Data regarding fetal cardiac diagnosis, extracardiac anomalies, nuchal translucency thickness and karyotype were noted. RESULTS: A total of 171 cases fulfilled our selection criteria. Of these, 69% were diagnosed with a perimembranous VSD and 31% with a muscular defect. The median gestational age at diagnosis was 21 + 6 (range, 12 + 0 to 37 + 3) weeks. Owing to severe extracardiac or genetic conditions, pregnancy resulted in intrauterine death or termination in 49% cases, and postnatal death occurred in 9% of cases. Seventy-two babies were liveborn, and were regarded as potential surgical candidates if hemodynamics suggested that surgery was indicated. Surgical closure of the VSD proved necessary in 50% of the patients with a perimembranous VSD and 13% of those with a muscular VSD. All patients operated on survived surgical repair. No karyotypic abnormalities were identified in fetuses with VSDs that had normal first-trimester screening and no other sonographic abnormalities. CONCLUSIONS: A high proportion of VSDs diagnosed during fetal life (29%) require postnatal surgical intervention. The assessment of hemodynamic significance from fetal echocardiography is imperfect. The presence of extracardiac abnormalities or abnormal results on first-trimester screening has a major impact on the incidence of karyotypic abnormalities in affected fetuses. This should inform discussions with parents about invasive testing.

Atopy phenotypes in the Childhood Asthma Prevention Study (CAPS) cohort and the relationship with allergic disease: clinical mechanisms in allergic disease.

BACKGROUND: Atopy in early life is heterogeneous in timing of onset, remission and persistence and in the nature of specific sensitization to allergens. However, this heterogeneity is not well characterized. OBJECTIVE: Our aim was to define longitudinal phenotypes of atopy between ages 1.5 and 8 years, and to assess the relationship of the atopy phenotypes to the risk of asthma, eczema and rhinitis at 8 years of age. METHODS: We used latent class analysis (LCA) to define atopy phenotypes using data from skin prick tests that were performed at 1.5, 3, 5 and 8 years in participants in the Childhood Asthma Prevention Study (CAPS). RESULTS: Four phenotypes were defined: late mixed inhalant sensitization; mixed food and inhalant sensitization; house dust mite (HDM) monosensitized; and no atopy. All three atopic phenotypes were associated with asthma, eczema and rhinitis, but the strongest association, particularly for asthma, was with the mixed food and inhalant sensitization phenotype. CONCLUSION & CLINICAL RELEVANCE: We have used a LCA model to define atopy phenotypes empirically. The finding of a strong association between the mixed food and inhalant sensitization class and the presence of asthma and poor asthma control at age 8 years implies that food sensitization in early life may be of greater significance for subsequent risk of asthma than previously thought.