Automatic Segmentation of the Fetus in 3D Magnetic Resonance Images Using Deep Learning: Accurate and Fast Fetal Volume Quantification for Clinical Use

Generalizable deep learning framework for 3D medical image segmentation using limited training data

Medical image segmentation is a critical component in a wide range of clinical applications, enabling the identification and delineation of anatomical structures. This study focuses on segmentation of anatomical structures for 3D printing, virtual surgery planning, and advanced visualization such as virtual or augmented reality. Manual segmentation methods are labor-intensive and can be subjective, leading to inter-observer variability. Machine learning algorithms, particularly deep learning models, have gained traction for automating the process and are now considered state-of-the-art. However, deep-learning methods typically demand large datasets for fine-tuning and powerful graphics cards, limiting their applicability in resource-constrained settings. In this paper we introduce a robust deep learning framework for 3D medical segmentation that achieves high performance across a range of medical segmentation tasks, even when trained on a small number of subjects. This approach overcomes the need for extensive data and heavy GPU resources, facilitating adoption within healthcare systems. The potential is exemplified through six different clinical applications involving orthopedics, orbital segmentation, mandible CT, cardiac CT, fetal MRI and lung CT. Notably, a small set of hyper-parameters and augmentation settings produced segmentations with an average Dice score of 92% (SD = ±0.06) across a diverse range of organs and tissues.

Segmentation and Estimation of Fetal Biometric Parameters using an Attention Gate Double U-Net with Guided Decoder Architecture

The fetus's health is evaluated with the biometric parameters obtained from the low-resolution ultrasound images. The accuracy of biometric parameters in existing protocols typically depends on conventional image processing approaches and hence, is prone to error. This study introduces the Attention Gate Double U-Net with Guided Decoder (ADU-GD) model specifically crafted for fetal biometric parameter prediction. The attention network and guided decoder are specifically designed to dynamically merge local features with their global dependencies, enhancing the precision of parameter estimation. The ADU-GD displays superior performance with Mean Absolute Error of 0.99 mm and segmentation accuracy of 99.1 % when benchmarked against the well-established models. The proposed model consistently achieved a high Dice index score of about 99.1 ± 0.8, with a minimal Hausdorff distance of about 1.01 ± 1.07 and a low Average Symmetric Surface Distance of about 0.25 ± 0.21, demonstrating the model's excellence. In a comprehensive evaluation, ADU-GD emerged as a frontrunner, outperforming existing deep-learning models such as Double U-Net, DeepLabv3, FCN-32s, PSPNet, SegNet, Trans U-Net, Swin U-Net, Mask-R2CNN, and RDHCformer models in terms of Mean Absolute Error for crucial fetal dimensions, including Head Circumference, Abdomen Circumference, Femur Length, and BiParietal Diameter. It achieved superior accuracy with MAE values of 2.2 mm, 2.6 mm, 0.6 mm, and 1.2 mm, respectively.

Deep learning-based segmentation of whole-body fetal MRI and fetal weight estimation: assessing performance, repeatability, and reproducibility

OBJECTIVES

To develop a deep-learning method for whole-body fetal segmentation based on MRI; to assess the method's repeatability, reproducibility, and accuracy; to create an MRI-based normal fetal weight growth chart; and to assess the sensitivity to detect fetuses with growth restriction (FGR).

METHODS

Retrospective data of 348 fetuses with gestational age (GA) of 19-39 weeks were included: 249 normal appropriate for GA (AGA), 19 FGR, and 80 Other (having various imaging abnormalities). A fetal whole-body segmentation model with a quality estimation module was developed and evaluated in 169 cases. The method was evaluated for its repeatability (repeated scans within the same scanner, n = 22), reproducibility (different scanners, n = 6), and accuracy (compared with birth weight, n = 7). A normal MRI-based growth chart was derived.

RESULTS

The method achieved a Dice = 0.973, absolute volume difference ratio (VDR) = 1.8% and VDR mean difference = 0.75% ([Formula: see text]: - 3.95%, 5.46), and high agreement with the gold standard. The method achieved a repeatability coefficient = 4.01%, ICC = 0.99, high reproducibility with a mean difference = 2.21% ([Formula: see text]: - 1.92%, 6.35%), and high accuracy with a mean difference between estimated fetal weight (EFW) and birth weight of - 0.39% ([Formula: see text]: - 8.23%, 7.45%). A normal growth chart (n = 246) was consistent with four ultrasound charts. EFW based on MRI correctly predicted birth-weight percentiles for all 18 fetuses ≤ 10thpercentile and for 14 out of 17 FGR fetuses below the 3rd percentile. Six fetuses referred to MRI as AGA were found to be < 3rd percentile.

CONCLUSIONS

The proposed method for automatic MRI-based EFW demonstrated high performance and sensitivity to identify FGR fetuses.

CLINICAL RELEVANCE STATEMENT

Results from this study support the use of the automatic fetal weight estimation method based on MRI for the assessment of fetal development and to detect fetuses at risk for growth restriction.

KEY POINTS

• An AI-based segmentation method with a quality assessment module for fetal weight estimation based on MRI was developed, achieving high repeatability, reproducibility, and accuracy. • An MRI-based fetal weight growth chart constructed from a large cohort of normal and appropriate gestational-age fetuses is proposed. • The method showed a high sensitivity for the diagnosis of small fetuses suspected of growth restriction.

Fetal cardiac magnetic resonance imaging of the descending aorta in suspected left-sided cardiac obstructions

Background

Severe left-sided cardiac obstructions are associated with high morbidity and mortality if not detected in time. The correct prenatal diagnosis of coarctation of the aorta (CoA) is difficult. Fetal cardiac magnetic resonance imaging (CMR) may improve the prenatal diagnosis of complex congenital heart defects. Flow measurements in the ascending aorta could aid in predicting postnatal CoA, but its accurate visualization is challenging.

Objectives

To compare the flow in the descending aorta (DAo) and umbilical vein (UV) in fetuses with suspected left-sided cardiac obstructions with and without the need for postnatal intervention and healthy controls by fetal phase-contrast CMR flow. A second objective was to determine if adding fetal CMR to echocardiography (echo) improves the fetal CoA diagnosis.

Methods

Prospective fetal CMR phase-contrast flow in the DAo and UV and echo studies were conducted between 2017 and 2022.

Results

A total of 46 fetuses with suspected left-sided cardiac obstructions [11 hypoplastic left heart syndrome (HLHS), five critical aortic stenosis (cAS), and 30 CoA] and five controls were included. Neonatal interventions for left-sided cardiac obstructions (n = 23) or comfort care (n = 1 with HLHS) were pursued in all 16 fetuses with suspected HLHS or cAS and in eight (27%) fetuses with true CoA. DAo or UV flow was not different in fetuses with and without need of intervention. However, DAo and UV flows were lower in fetuses with either retrograde isthmic systolic flow [DAo flow 253 (72) vs. 261 (97) ml/kg/min, p = 0.035; UV flow 113 (75) vs. 161 (81) ml/kg/min, p = 0.04] or with suspected CoA and restrictive atrial septum [DAo flow 200 (71) vs. 268 (94) ml/kg/min, p = 0.04; UV flow 89 vs. 159 (76) ml/kg/min, p = 0.04] as well as in those without these changes. Adding fetal CMR to fetal echo predictors for postnatal CoA did not improve the diagnosis of CoA.

Conclusion

Fetal CMR-derived DAo and UV flow measurements do not improve the prenatal diagnosis of left-sided cardiac obstructions, but they could be important in identifying fetuses with a more severe decrease in blood flow across the left side of the heart. The physiological explanation may be a markedly decreased left ventricular cardiac output with subsequent retrograde systolic isthmic flow and decreased total DAo flow.