Search

COVID-19 is clinically characterised by fever, cough, and dyspnoea. Symptoms affecting other organ systems have been reported. However, it is the clinical associations of different patterns of symptoms which influence diagnostic and therapeutic decision-making. In this study, we applied clustering techniques to a large prospective cohort of hospitalised patients with COVID-19 to identify clinically meaningful sub-phenotypes. We obtained structured clinical data on 59,011 patients in the UK (the ISARIC Coronavirus Clinical Characterisation Consortium, 4C) and used a principled, unsupervised clustering approach to partition the first 25,477 cases according to symptoms reported at recruitment. We validated our findings in a second group of 33,534 cases recruited to ISARIC-4C, and in 4,445 cases recruited to a separate study of community cases. Unsupervised clustering identified distinct sub-phenotypes. First, a core symptom set of fever, cough, and dyspnoea, which co-occurred with additional symptoms in three further patterns: fatigue and confusion, diarrhoea and vomiting, or productive cough. Presentations with a single reported symptom of dyspnoea or confusion were also identified, alongside a sub-phenotype of patients reporting few or no symptoms. Patients presenting with gastrointestinal symptoms were more commonly female, had a longer duration of symptoms before presentation, and had lower 30-day mortality. Patients presenting with confusion, with or without core symptoms, were older and had a higher unadjusted mortality. Symptom sub-phenotypes were highly consistent in replication analysis within the ISARIC-4C study. Similar patterns were externally verified in patients from a study of self-reported symptoms of mild disease. The large scale of the ISARIC-4C study enabled robust, granular discovery and replication. Clinical interpretation is necessary to determine which of these observations have practical utility. We propose that four sub-phenotypes are usefully distinct from the core symptom group: gastro-intestinal disease, productive cough, confusion, and pauci-symptomatic presentations. Importantly, each is associated with an in-hospital mortality which differs from that of patients with core symptoms.

ObjectiveProgranulin-related frontotemporal dementia (FTD-GRN) is a fast progressive disease. Modelling the cascade of multimodal biomarker changes aids in understanding the aetiology of this disease and enables monitoring of individual mutation carriers. In this cross-sectional study, we estimated the temporal cascade of biomarker changes for FTD-GRN, in a data-driven way.MethodsWe included 56 presymptomatic and 35 symptomatic GRN mutation carriers, and 35 healthy non-carriers. Selected biomarkers were neurofilament light chain (NfL), grey matter volume, white matter microstructure and cognitive domains. We used discriminative event-based modelling to infer the cascade of biomarker changes in FTD-GRN and estimated individual disease severity through cross-validation. We derived the biomarker cascades in non-fluent variant primary progressive aphasia (nfvPPA) and behavioural variant FTD (bvFTD) to understand the differences between these phenotypes.ResultsLanguage functioning and NfL were the earliest abnormal biomarkers in FTD-GRN. White matter tracts were affected before grey matter volume, and the left hemisphere degenerated before the right. Based on individual disease severities, presymptomatic carriers could be delineated from symptomatic carriers with a sensitivity of 100% and specificity of 96.1%. The estimated disease severity strongly correlated with functional severity in nfvPPA, but not in bvFTD. In addition, the biomarker cascade in bvFTD showed more uncertainty than nfvPPA.ConclusionDegeneration of axons and language deficits are indicated to be the earliest biomarkers in FTD-GRN, with bvFTD being more heterogeneous in disease progression than nfvPPA. Our data-driven model could help identify presymptomatic GRN mutation carriers at risk of conversion to the clinical stage.

OBJECTIVES: the MOMS randomized trial has demonstrated that prenatal spina bifida aperta (SBA) repair via open approach is safe and effective for both mother and fetus, yet half of the infants have incomplete reversal of hindbrain herniation and no improvement in neuromotor function. One contributing factor may be the incompleteness of the neurosurgical repair causing persistent in utero leakage of cerebrospinal fluid (CSF) and exposure of the spinal cord to the amniotic fluid until birth. We aim to investigate the neurostructural and neurofunctional efficacy of the watertightness of prenatal SBA repair. METHODS: a superiority study was conducted in the validated SBA fetal lamb model and powered (n=7 per group). Outcomes of lambs undergoing watertight or non-watertight multilayer repairs through an open approach were compared to those in unrepaired SBA lambs at delivery (term=145 days). At ~75-day fetal lambs underwent standardized induction of lumbar SBA. At ~100 days, they were assigned to one of the three groups and subsequently to either watertight or non-watertight layered repair group based on an intraoperative watertightness test using subcutaneous fluoresceine injection. Finally, at 1-2 postnatal days, we assessed reversal of hindbrain herniation on Magnetic Resonance Imaging (MRI) as primary outcome. Secondary proxies of neuroprotection were CSF leakage at the repair site; hindlimb motor function based on joint movement score, locomotor grade and Motor Evoked Potentials (MEP); neuroprotection score encompassing live birth, hindbrain herniation reversal, absent CSF leakage and joint score =9/15; and brain and spinal cord histology and immunohistochemistry. As the watertightness test is not clinically usable, we developed a potential surrogate intraoperative quality score and assessed its relationship with improved outcomes. This four-point scoring system is based on visual assessment of the quality of the skin repair (suture inter-run distance =3mm, absence of tear and ischemia). RESULTS: Compared to unrepaired lambs, watertight repair was neuroprotective in 5/7 lambs (neuroprotection score of 4/4), as evidenced by a 100% reversal of hindbrain herniation on MRI; lower CSF leakage rate (14%); better hindlimb motor function evidenced by higher joint movement score, locomotor grade, and MEP area-under-the-curve and peak-to-peak amplitude; higher neuronal density in the hippocampus and corpus callosum; and higher reactive astrogliosis at the SBA lesion epicenter. Conversely, non-watertight SBA repair did not achieve the same level of neuroprotection (1/7) due to non-significant 86% hindbrain herniation reversal, no motor function improvement, high CSF leakage (43%), low brain neuron count in both regions and low spinal astroglial cell area at the epicenter. Like watertight layered repair, a high quality score (=2/3) was associated with improved outcomes yet watertightness test and quality score could not be used interchangeably due to results discrepancies. CONCLUSIONS: A watertight layered fetal SBA repair improves brain and spinal cord structure and function in the fetal lamb model. This translational research has important clinical implication and neurosurgical technique that achieves watertightness should be adopted in all fetal centers to improve neuroprotection. Future clinical studies could assess whether the high quality of the repair correlates with clinical outcomes for neuroprotection. This article is protected by copyright. All rights reserved.

Objective To assess the safety of Partial-Amniotic-Insufflation-of-heated-humidified-CO2 (hPACI) during fetoscopic spina bifida repair (fSB-repair). Method A simulated fSB-repair through an exteriorized uterus under hPACI was performed in 100-day fetal lambs (term = 145 days) under a laboratory anesthesia protocol (n = 5; group 1) which is known to induce maternal-fetal acidosis and hypercapnia. Since these may not occur clinically, we applied a clinical anesthesia protocol (n = 5; group 2), keeping maternal parameters within physiological conditions, that is, controlled maternal arterial carbon dioxide (CO2) pressure (pCO(2) = 30 mmHg), blood pressure (>= 67 mmHg), and temperature (37.1-39.8 degrees C). Our superiority study used fetal pH as the primary outcome. Results Compared to group 1, controlled anesthesia normalized fetal pH (7.23 +/- 0.02 vs. 7.36 +/- 0.02, p < 0.001), pCO(2) (70.0 +/- 9.1 vs. 43.0 +/- 1.0 mmHg, p = 0.011) and bicarbonate (27.8 +/- 1.1 vs. 24.0 +/- 0.9 mmol/L, p = 0.071) at baseline. It kept them within clinically acceptable limits (pH >= 7.23, pCO(2) = 120 min of hPACI as opposed to

This paper describes our method for our participation in the FeTA challenge2021 (team name: TRABIT). The performance of convolutional neural networks for medical image segmentation is thought to correlate positively with the number of training data. The FeTA challenge does not restrict participants to using only the provided training data but also allows for using other publicly available sources. Yet, open access fetal brain data remains limited. An advantageous strategy could thus be to expand the training data to cover broader perinatal brain imaging sources. Perinatal brain MRIs, other than the FeTA challenge data, that are currently publicly available, span normal and pathological fetal atlases as well as neonatal scans. However, perinatal brain MRIs segmented in different datasets typically come with different annotation protocols. This makes it challenging to combine those datasets to train a deep neural network. We recently proposed a family of loss functions, the label-set loss functions, for partially supervised learning. Label-set loss functions allow to train deep neural networks with partially segmented images, i.e. segmentations in which some classes may be grouped into super-classes. We propose to use label-set loss functions to improve the segmentation performance of a state-of-the-art deep learning pipeline for multi-class fetal brain segmentation by merging several publicly available datasets. To promote generalisability, our approach does not introduce any additional hyper-parameters tuning., Accepted as a poster at the MICCAI 2021 Perinatal, Preterm and Paediatric Image Analysis (PIPPI) workshop

The performance of deep neural networks typically increases with the number of training images. However, not all images have the same importance towards improved performance and robustness. In fetal brain MRI, abnormalities exacerbate the variability of the developing brain anatomy compared to non-pathological cases. A small number of abnormal cases, as is typically available in clinical datasets used for training, are unlikely to fairly represent the rich variability of abnormal developing brains. This leads machine learning systems trained by maximizing the average performance to be biased toward non-pathological cases. This problem was recently referred to as hidden stratification. To be suited for clinical use, automatic segmentation methods need to reliably achieve high-quality segmentation outcomes also for pathological cases. In this paper, we show that the state-of-the-art deep learning pipeline nnU-Net has difficulties to generalize to unseen abnormal cases. To mitigate this problem, we propose to train a deep neural network to minimize a percentile of the distribution of per-volume loss over the dataset. We show that this can be achieved by using Distributionally Robust Optimization (DRO). DRO automatically reweights the training samples with lower performance, encouraging nnU-Net to perform more consistently on all cases. We validated our approach using a dataset of 368 fetal brain T2w MRIs, including 124 MRIs of open spina bifida cases and 51 MRIs of cases with other severe abnormalities of brain development., Accepted at the MICCAI 2021 Perinatal, Preterm and Paediatric Image Analysis (PIPPI) workshop. arXiv admin note: substantial text overlap with arXiv:2001.02658

Deep neural networks have increased the accuracy of automatic segmentation, however, their accuracy depends on the availability of a large number of fully segmented images. Methods to train deep neural networks using images for which some, but not all, regions of interest are segmented are necessary to make better use of partially annotated datasets. In this paper, we propose the first axiomatic definition of label-set loss functions that are the loss functions that can handle partially segmented images. We prove that there is one and only one method to convert a classical loss function for fully segmented images into a proper label-set loss function. Our theory also allows us to define the leaf-Dice loss, a label-set generalization of the Dice loss particularly suited for partial supervision with only missing labels. Using the leaf-Dice loss, we set a new state of the art in partially supervised learning for fetal brain 3D MRI segmentation. We achieve a deep neural network able to segment white matter, ventricles, cerebellum, extra-ventricular CSF, cortical gray matter, deep gray matter, brainstem, and corpus callosum based on fetal brain 3D MRI of anatomically normal fetuses or with open spina bifida. Our implementation of the proposed label-set loss functions is available at https://github.com/LucasFidon/label-set-loss-functions, Accepted at MICCAI 2021

Detailed analysis of seizure semiology, the symptoms and signs which occur during a seizure, is critical for management of epilepsy patients. Inter-rater reliability using qualitative visual analysis is often poor for semiological features. Therefore, automatic and quantitative analysis of video-recorded seizures is needed for objective assessment. We present GESTURES, a novel architecture combining convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to learn deep representations of arbitrarily long videos of epileptic seizures. We use a spatiotemporal CNN (STCNN) pre-trained on large human action recognition (HAR) datasets to extract features from short snippets (approx. 0.5 s) sampled from seizure videos. We then train an RNN to learn seizure-level representations from the sequence of features. We curated a dataset of seizure videos from 68 patients and evaluated GESTURES on its ability to classify seizures into focal onset seizures (FOSs) (N = 106) vs. focal to bilateral tonic-clonic seizures (TCSs) (N = 77), obtaining an accuracy of 98.9% using bidirectional long short-term memory (BLSTM) units. We demonstrate that an STCNN trained on a HAR dataset can be used in combination with an RNN to accurately represent arbitrarily long videos of seizures. GESTURES can provide accurate seizure classification by modeling sequences of semiologies., Accepted at the 24th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2021)

While convolutional neural networks (CNNs) trained by back-propagation have seen unprecedented success at semantic segmentation tasks, they are known to struggle on out-of-distribution data. Markov random fields (MRFs) on the other hand, encode simpler distributions over labels that, although less flexible than UNets, are less prone to over-fitting. In this paper, we propose to fuse both strategies by computing the product of distributions of a UNet and an MRF. As this product is intractable, we solve for an approximate distribution using an iterative mean-field approach. The resulting MRF-UNet is trained jointly by back-propagation. Compared to other works using conditional random fields (CRFs), the MRF has no dependency on the imaging data, which should allow for less over-fitting. We show on 3D neuroimaging data that this novel network improves generalisation to out-of-distribution samples. Furthermore, it allows the overall number of parameters to be reduced while preserving high accuracy. These results suggest that a classic MRF smoothness prior can allow for less over-fitting when principally integrated into a CNN model. Our implementation is available at https://github.com/balbasty/nitorch., Accepted at MIDL 2021

A novel phantom-imaging platform for automated and high precision imaging of the American College of Radiology (ACR) PET phantom is proposed. The platform facilitates the generation of an accurate μ-map for PET/MR systems with a robust alignment based on two-stage image registration using specifically designed PET templates. The automated analysis of PET images uses a set of granular composite volume of interest (VOI) templates in a 0.5 mm resolution grid for sampling of the system response to the insert step functions. The impact of the activity outside the field of view (FOV) was evaluated using two acquisitions of 30 minutes each, with and without the activity outside the FOV. Iterative image reconstruction was employed with and without modelled shift-invariant point spread function (PSF) and varying ordered subsets expectation maximisation (OSEM) iterations. Uncertainty analysis of all image-derived statistics was performed using bootstrap resampling of the list-mode data. We found that the activity outside the FOV can adversely affect the imaging planes close to the edge of the axial FOV, reducing the contrast, background uniformity and overall quantitative accuracy. The PSF had a positive impact on contrast recovery (although it slows convergence). The proposed platform may be helpful in a more informative evaluation of PET systems and image reconstruction methods.

Open fetal surgery for spina bifida (SB) is safe and effective yet invasive. The growing interest in fetoscopic SB repair (fSB-repair) prompts the need for appropriate training. We aimed to develop and validate a high-fidelity training model for fSB-repair. fSB-repair was simulated in the abdominal cavity and on the stomach of adult rabbits. Laparoscopic fetal surgeons served either as novices (n = 2) or experts (n = 3) based on their experience. Technical performance was evaluated using competency Cumulative Sum (CUSUM) analysis and the group splitting method. Main outcome measure for CUSUM competency was a composite binary outcome for surgical success, i.e. watertight repair, operation time = 180 min and Objective-Structured-Assessment-of-Technical-Skills (OSATS) score = 18/25. Construct validity was first confirmed since competency levels of novices and experts during their six first cases using both methods were significantly different. Criterion validity was also established as 33 consecutive procedures were needed for novices to reach competency using learning curve CUSUM, which is a number comparable to that of clinical fSB-repair. Finally, we surveyed expert fetal surgeons worldwide to assess face and content validity. Respondents (26/49; 53%) confirmed it with = 71% of scores for overall realism = 4/7 and usefulness = 3/5. We propose to use our high-fidelity model to determine and shorten the learning curve of laparoscopic fetal surgeons and retain operative skills.

Background: Structural cortical networks (SCNs) represent patterns of coordinated morphological modifications in cortical areas, and they present the advantage of being extracted from previously acquired clinical magnetic resonance imaging (MRI) scans. SCNs have shown pathophysiological changes in many brain disorders, including multiple sclerosis. Objective: To investigate alterations of SCNs at the individual level in patients with clinically isolated syndrome (CIS), thereby assessing their clinical relevance. Methods: We analyzed baseline data collected in a prospective multicenter (MAGNIMS) study. CIS patients ( n = 60) and healthy controls ( n = 38) underwent high-resolution 3T MRI. Measures of disability and cognitive processing were obtained for patients. Single-subject SCNs were extracted from brain 3D-T1 weighted sequences; global and local network parameters were computed. Results: Compared to healthy controls, CIS patients showed altered small-world topology, an efficient network organization combining dense local clustering with relatively few long-distance connections. These disruptions were worse for patients with higher lesion load and worse cognitive processing speed. Alterations of centrality measures and clustering of connections were observed in specific cortical areas in CIS patients when compared with healthy controls. Conclusion: Our study indicates that SCNs can be used to demonstrate clinically relevant alterations of connectivity in CIS.