A Dempster-Shafer Approach to Trustworthy AI With Application to Fetal Brain MRI Segmentation

Deep learning models for medical image segmentation can fail unexpectedly and spectacularly for pathological cases and images acquired at different centers than training images, with labeling errors that violate expert knowledge. Such errors undermine the trustworthiness of deep learning models for medical image segmentation. Mechanisms for detecting and correcting such failures are essential for safely translating this technology into clinics and are likely to be a requirement of future regulations on artificial intelligence (AI). In this work, we propose a trustworthy AI theoretical framework and a practical system that can augment any backbone AI system using a fallback method and a fail-safe mechanism based on Dempster-Shafer theory. Our approach relies on an actionable definition of trustworthy AI. Our method automatically discards the voxel-level labeling predicted by the backbone AI that violate expert knowledge and relies on a fallback for those voxels. We demonstrate the effectiveness of the proposed trustworthy AI approach on the largest reported annotated dataset of fetal MRI consisting of 540 manually annotated fetal brain 3D T2w MRIs from 13 centers. Our trustworthy AI method improves the robustness of four backbone AI models for fetal brain MRIs acquired across various centers and for fetuses with various brain abnormalities.

Long-term pulmonary and neurodevelopmental impairment in a fetal growth restriction rabbit model

Fetal growth restriction (FGR) remains one of the main obstetrical problems worldwide, with consequences beyond perinatal life. Animal models with developmental and structural similarities to the human are essential to understand FGR long-term consequences and design novel therapeutic strategies aimed at preventing or ameliorating them. Herein, we described the long-term consequences of FGR in pulmonary function, structure, and gene expression, and characterized neurodevelopmental sequelae up to preadolescence in a rabbit model. FGR was induced at gestational day 25 by surgically reducing placental blood supply in one uterine horn, leaving the contralateral horn as internal control. Neonatal rabbits born near term were assigned to foster care in mixed groups until postnatal day (PND) 21. At that time, one group underwent pulmonary biomechanical testing followed by lung morphometry and gene expression analysis. A second group underwent longitudinal neurobehavioral assessment until PND 60 followed by brain harvesting for multiregional oligodendrocyte and microglia quantification. FGR was associated with impaired pulmonary function and lung development at PND 21. FGR rabbits had higher respiratory resistance and altered parenchymal biomechanical properties in the lungs. FGR lungs presented thicker alveolar septal walls and reduced alveolar space. Furthermore, the airway smooth muscle content was increased, and the tunica media of the intra-acinar pulmonary arteries was thicker. In addition, FGR was associated with anxiety-like behavior, impaired memory and attention, and lower oligodendrocyte proportion in the frontal cortex and white matter. In conclusion, we documented and characterized the detrimental pulmonary function and structural changes after FGR, independent of prematurity, and beyond the neonatal period for the first time in the rabbit model, and describe the oligodendrocyte alteration in pre-adolescent rabbit brains. This characterization will allow researchers to develop and test therapies to treat FGR and prevent its sequelae.

Duration of fetoscopic spina bifida repair does not affect the central nervous system in fetal lambs

BACKGROUND

Prenatal spina bifida aperta repair improves neurologic outcomes yet comes with a significant risk of prematurity and uterine scar-related complications. To reduce such complications, different fetoscopic techniques, for example, with varying numbers of ports, are being explored. This has an effect on the duration of the procedure, potentially affecting central nervous system development. Both the condition and anesthesia can affect the central nervous system, particularly the hippocampus, a region crucial for prospective and episodic memory. Previous animal studies have shown the potential influence of anesthesia, premature delivery, and maternal surgery during pregnancy on this area.

OBJECTIVE

This study aimed to compare the effects of 2- vs 3-port fetoscopic spina bifida aperta repair in the fetal lamb model using neuron count of the hippocampus as the primary outcome.

STUDY DESIGN

Based on the hippocampal neuron count from previous lamb experiments, we calculated that we required 5 animals per group to achieve a statistical power of ≥ 80%. A spina bifida aperta defect was developed in fetal lambs at 75 days of gestation (term: 145 days). At 100 days, fetuses underwent either a 2-port or 3-port fetoscopic repair. At 143 days, all surviving fetuses were delivered by cesarean delivery, anesthetized, and transcardially perfused with a mixture of formaldehyde and gadolinium. Next, they underwent neonatal brain and spine magnetic resonance imaging after which these organs were harvested for histology. Hippocampus, frontal cortex, caudate nucleus, and cerebellum samples were immunostained to identify neurons, astrocytes, microglia, and markers associated with cell proliferation, myelination, and synapses. The degree of hindbrain herniation and the ventricular diameter were measured on magnetic resonance images and volumes of relevant brain and medulla areas were segmented.

RESULTS

Both treatment groups included 5 fetuses and 9 unoperated littermates served as normal controls. The durations for both skin-to-skin (341±31 vs 287±40 minutes; P=.04) and fetal surgery (183±30 vs 128±22; P=.01) were longer for the 2-port approach than for the 3-port approach. There was no significant difference in neuron density in the hippocampus, frontal cortex, and cerebellum. In the caudate nucleus, the neuron count was higher in the 2-port group (965±156 vs 767±92 neurons/mm2; P=.04). There were neither differences in proliferation, astrogliosis, synaptophysin, or myelin. The tip of the cerebellar vermis was closer to the foramen magnum in animals undergoing the 2-port approach than in animals undergoing the 3-port approach (-0.72±0.67 vs -2.47±0.91 mm; P=.009). There was no significant difference in the ratio of the hippocampus, caudate nucleus, or cerebellar volume to body weight. For the spine, no difference was noted in spine volume-to-body weight ratio for the lower (L1-L2), middle (L3-L4), and higher (L5-L6) levels. Compared with controls, in repaired animals, the cerebellar vermis tip laid closer to the foramen magnum, parietal ventricles were enlarged, and medulla volumes were reduced.

CONCLUSION

In the experimental spina bifida fetal lamb model, a 2-port repair took 40% longer than a 3-port repair. However, there was no indication of any relevant morphologic differences in the fetal brain.

Effects of cumulative duration of repeated anaesthesia exposure on foetal brain development in the ovine model

OBJECTIVE

Anaesthesia is required in 0.4-1% of pregnant women, and prolonged and repeated exposures to anaesthesia may be required. It is unknown whether these exposures may result in foetal neurotoxicity in humans. As sheep have a gestation comparable to that of humans, the objective of this study was to analyse the neurodevelopmental outcome of ovine foetuses that had been exposed in utero to repeated and prolonged anaesthesia.

DESIGN

Randomized controlled preclinical study.

SETTING

Anaesthesia for non-obstetric surgery during pregnancy.

ANIMALS

Twenty-four healthy pregnant Swifter ewes.

INTERVENTIONS

The ewes were randomized to no anaesthesia exposure (control-group), single exposure (at gestational age 68-70 days), or repeated exposure (at gestational age 68-70 days and 96-98 days) to 2.5 h of sevoflurane anaesthesia and maternal laparotomy. All lambs were delivered at approximately term gestation (gestational age: 140-143 days).

MEASUREMENTS

The primary outcome was neuron density in the frontal cortex 24 h after birth for the control-group versus the repeated-exposure-group. Key secondary outcome was the time needed to achieve the milestone of standing. Secondary outcomes included other neurobehavioural assessments (e.g., motoric milestones) and histological parameters quantified in multiple brain regions (neuron density, total cell density, proliferation, inflammation, synaptogenesis, astrocytes and myelination).

MAIN RESULTS

Neuron density in the frontal cortex did not differ between groups (mean ± standard deviation: control-group: 403 ± 39, single-exposure group: 436 ± 23 and repeated-exposure-group: 403 ± 40 neurons/mm2, control-group versus repeated-exposure-group: p = 0.986, control-group versus single-exposure-group: p = 0.097). No significant difference was observed for the time needed to achieve the milestone of standing. Only very limited differences were observed for other histological outcome parameters and neurobehavioural assessments.

CONCLUSIONS

There is no evidence for foetal neuronal injury or neurobehavioural impairments after a cumulative duration of 5 h repetitive prenatal anaesthesia in sheep.

Brain development is altered in rabbit fetuses with congenital diaphragmatic hernia

INTRODUCTION

Children with congenital diaphragmatic hernia (CDH) are at risk for neurodevelopmental delay. Some changes are already present prenatally. Herein, we further examined how the brain develops in fetal rabbits with surgically created DH.

METHODS

Two fetuses underwent surgical DH creation on day 23 (term = d31). DH pups and littermate controls were harvested at term. Ten DH pups and 11 controls underwent transcardial perfusion for brain fixation and measurement of brain volume, brain folding, neuron and synaptic density, pre-oligodendrocyte count, proliferation, and vascularization. Twelve other DH and 11 controls had echocardiographic assessment of cardiac output and aortic and cerebral blood flow, magnetic resonance imaging (9.4 T) for cerebral volumetry, and molecular assessment of vascularization markers.

RESULTS

DH pups had lower lung-to-body weight ratio (1.3 ± 0.3 vs. 2.4 ± 0.3%; p < 0.0001) and lower heart-to-body weight ratio (0.007 ± 0.001 vs. 0.009 ± 0.001; p = 0.0006) but comparable body weight and brain-to-body weight ratio. DH pups had a lower left ventricular ejection fraction, aortic and cerebral blood flow (39 ± 8 vs. 54 ± 15 mm/beat; p = 0.03) as compared to controls but similar left cardiac ventricular morphology. Fetal DH-brains were similar in volume but the cerebellum was less folded (perimeter/surface area: 25.5 ± 1.5 vs. 26.8 ± 1.2; p = 0.049). Furthermore, DH brains had a thinner cortex (143 ± 9 vs. 156 ± 13 μm; p = 0.02). Neuron densities in the white matter were higher in DH fetuses (124 ± 18 vs. 104 ± 14; p = 0.01) with comparable proliferation rates. Pre-oligodendrocyte count was lower, coinciding with the lower endothelial cell count.

CONCLUSION

Rabbits with DH had altered brain development compared to controls prenatally, indicating that brain development is already altered prenatally in CDH.

Longitudinal MRI Evaluation of Brain Development in Fetuses with Congenital Diaphragmatic Hernia around the Time of Fetal Endotracheal Occlusion

BACKGROUND AND PURPOSE

Congenital diaphragmatic hernia is associated with high mortality and morbidity, including evidence suggesting neurodevelopmental comorbidities after birth. The aim of this study was to document longitudinal changes in brain biometry and the cortical folding pattern in fetuses with congenital diaphragmatic hernia compared with healthy fetuses.

MATERIALS AND METHODS

This is a retrospective cohort study including fetuses with isolated congenital diaphragmatic hernia between January 2007 and May 2019, with at least 2 MR imaging examinations. For controls, we used images from fetuses who underwent MR imaging for an unrelated condition that did not compromise fetal brain development and fetuses from healthy pregnant women. Biometric measurements and 3D segmentations of brain structures were used as well as qualitative and quantitative grading of the supratentorial brain. Brain development was correlated with disease-severity markers.

RESULTS

Forty-two fetuses were included, with a mean gestational age at first MR imaging of 28.0 (SD, 2.1) weeks and 33.2 (SD, 1.3) weeks at the second imaging. The mean gestational age in controls was 30.7 (SD, 4.2) weeks. At 28 weeks, fetuses with congenital diaphragmatic hernia had abnormal qualitative and quantitative maturation, more extra-axial fluid, and larger total skull volume. By 33 weeks, qualitative grading scores were still abnormal, but quantitative scoring was in the normal range. In contrast, the extra-axial fluid volume remained abnormal with increased ventricular volume. Normal brain parenchymal volumes were found.

CONCLUSIONS

Brain development in fetuses with congenital diaphragmatic hernia around 28 weeks appears to be delayed. This feature is less prominent at 33 weeks. At this stage, there was also an increase in ventricular and extra-axial space volume.

Fetal surgery has no additional effect to general anesthesia on brain development in neonatal rabbits

BACKGROUND

Fetal surgery is part of modern fetal medicine, and some procedures, such as fetal spina bifida repair, are performed under general anesthesia. Fetuses are operated on in a time window when the developing brain is extremely vulnerable to external, potentially harmful factors. To date, little is known about the effect of fetal surgery on fetal brain development.

OBJECTIVE

This study aimed to assess the effect of fetal surgery on the developing fetal brain in the rabbit model.

STUDY DESIGN

This was a randomized, sham-controlled study in time-mated pregnant does at 28 days' gestation (term, 31 days), which corresponds to the start of the peak of brain development and end of the second trimester of pregnancy in humans. We included 4 different groups in this experiment: no-surgery, general anesthesia, general anesthesia+hysterotomy, and general anesthesia+fetal surgery. In 11 does, anesthesia was induced using propofol and maintained for 75 minutes with 3.6 vol% (4% is the equivalent of 1 minimum alveolar concentration) sevoflurane. Maternal blood pressure, heart rate, oxygen saturation, temperature, end-tidal CO2 were continuously monitored. For each operated doe, 6 fetuses were part of the experiment. Randomization determined which cornual sac and what opposing third sac were assigned to fetal surgery: hysterotomy, fetal injection (atropine, fentanyl, and cisatracurium), fetal skin incision, and suturing. Only hysterotomy was performed on the opposing cornual and third amniotic sacs of the does. The fetus in these experimental sacs was used as internal unmanipulated control (general anesthesia). All fetuses (n=38) from unmanipulated does (n=4) served as external controls (no-surgery). At term, the does were delivered by cesarean delivery under ketamine-medetomidine sedation and local anesthesia. The pups underwent standardized motoric and sensory neurologic testing on day 1 followed by euthanasia and brain harvesting for histologic assessment of neurons, synapses, proliferation, and glial cells.

RESULTS

Maternal vital signs were stable during surgery. Survival was similar in the 4 groups (75%-94%), and brain-to-body weight ratio was comparable; only the no-surgery pups had a higher brain weight. On postnatal day 1, the pups in the 4 groups had a comparable neurobehavioral outcome on both motoric and sensory testing. In the prefrontal cortex, no-surgery pups had significantly higher neuron density than pups who underwent maternal surgery, but there was no difference among pups that underwent general anesthesia, hysterotomy, or fetal surgery. The measurements of proliferation had a similar outcome: a higher proliferation rate in the prefrontal cortex of no-surgery pups. Moreover, synaptic density values were higher in the no-surgery pups, but there was no difference observed among pups who underwent general anesthesia, hysterotomy, and fetal surgery. Lastly, there was no difference in gliosis among the 4 groups.

CONCLUSION

In rabbits, fetal surgery through hysterotomy under maternal general anesthesia did not affect brain development, in addition to the effects of general anesthesia per se.

Prenatal cerebellar growth is altered in congenital diaphragmatic hernia on ultrasound

OBJECTIVE

Children with congenital diaphragmatic hernia (CDH) are at risk for neurodevelopmental delay. Herein we report on prenatal changes in biometry and brain perfusion in fetuses with isolated CDH.

STUDY DESIGN

This retrospective study evaluated fetuses with isolated, left-sided CDH in three European referral centers. Abdominal circumference (AC), femur length (FL), head circumference (HC), transcerebellar diameter (TCD), middle cerebral artery (MCA) Doppler, and ventricular width (VW) were assessed during four gestational periods (<24 weeks, 25-28 weeks, 29-32 weeks, >33 weeks). Z-scores were calculated, and growth curves were created based on longitudinal data.

RESULTS

In 367 fetuses, HC, AC and FL were within normal ranges throughout gestation. The TCD diminished with advancing gestational age to fall below the fifth percentile after 32 weeks. A less pronounced but similar trend was seen in VW. The peak systolic velocity of the MCA was consistently approximately 10% lower than normal. Disease severity was correlated to TCD (p = 0.002) and MCA doppler values (p = 0.002). There were no differences between fetuses treated with FETO and those managed expectantly.

CONCLUSION

Fetuses with isolated left-sided CDH have a small cerebellum and reduced MCA peak systolic velocity. Follow up studies are necessary to determine the impact of these changes on neurodevelopment.

Effects of Maternal Abdominal Surgery on Fetal Brain Development in the Rabbit Model

INTRODUCTION

Anesthesia during pregnancy can impair fetal neurodevelopment, but effects of surgery remain unknown. The aim is to investigate effects of abdominal surgery on fetal brain development. Hypothesis is that surgery impairs outcome.

METHODS

Pregnant rabbits were randomized at 28 days of gestation to 2 h of general anesthesia (sevoflurane group, n = 6) or to anesthesia plus laparoscopic appendectomy (surgery group, n = 13). On postnatal day 1, neurobehavior of pups was assessed and brains harvested. Primary outcome was neuron density in the frontal cortex, and secondary outcomes included neurobehavioral assessment and other histological parameters.

RESULTS

Fetal survival was lower in the surgery group: 54 versus 100% litters alive at birth (p = 0.0442). In alive litters, pup survival until harvesting was 50 versus 69% (p = 0.0352). No differences were observed for primary outcome (p = 0.5114) for surviving pups. Neuron densities were significantly lower in the surgery group in the caudate nucleus (p = 0.0180), but not different in other regions. No differences were observed for secondary outcomes. Conclusions did not change after adjustment for mortality.

CONCLUSION

Abdominal surgery in pregnant rabbits at a gestational age corresponding to the end of human second trimester results in limited neurohistological changes but not in neurobehavioral impairments. High intrauterine mortality limits translation to clinical scenario, where fetal mortality is close to zero.