Publications
1021 results found
De Marvao A, McGurk K, Zheng S, et al., 2021, Outcomes and phenotypic expression of rare variants in hypertrophic cardiomyopathy genes in over 200,000 adults, ESC Congress 2021, Publisher: European Society of Cardiology, Pages: 1731-1731, ISSN: 0195-668X
BackgroundHypertrophic cardiomyopathy (HCM) is caused by rare variants in sarcomere-encoding genes, but little is known about the clinical significance of these variants in the general population.PurposeTo determine the population prevalence of HCM-associated sarcomeric variants, characterise their phenotypic manifestations, estimate penetrance, and identify associations between sarcomeric variants and clinical outcomes, we performed an observational study of 218,813 adults in the UK Biobank (UKBB), of whom 200,584 have whole exome sequencing (WES).MethodsWe carried out an integrated analysis of WES and cardiac magnetic resonance (CMR) imaging in UK Biobank participants stratified by sarcomere-encoding variant status. Computer vision techniques were used to automatically segment the four chambers of the heart (Figure 1). Cardiac motion analysis was used to derive strain and strain rates. Regional analysis of left ventricular wall thickness was performed using three-dimensional modelling of these segmentations.ResultsMedian age at recruitment was 58 (IQR 50–63 years), and participants were followed up for a median of 10.8 years (IQR 9.9–11.6 years) with a total of 19,507 primary clinical events reported.The prevalence of rare variants (allele frequency <0.ehab724.17314) in HCM-associated sarcomere-encoding genes in 200,584 participants was 2.9% (n=5,727; 1 in 35), and the prevalence of pathogenic or likely pathogenic variants (SARC-P/LP) was 0.24% (n=474, 1 in 423).SARC-P/LP variants were associated with increased risk of death or major adverse cardiac events (MACE) compared to controls (HR 1.68, 95% CI 1.37–2.06, p<0.001), mainly due to heart failure endpoints (Figure 2: cumulative hazard curves with zoomed plots for lifetime risk of A) death and MACE or B) heart failure, stratified by genotype; genotype negative (SARC-NEG), carriers of indeterminate sarcomeric variants (SARC-IND) or SARC-P/LP; C) Forest plot of comparative lifetime risk of c
Makowski MR, Bressem KK, Franz L, et al., 2021, De Novo Radiomics Approach Using Image Augmentation and Features From T1 Mapping to Predict Gleason Scores in Prostate Cancer., Invest Radiol, Vol: 56, Pages: 661-668
OBJECTIVES: The aims of this study were to discriminate among prostate cancers (PCa's) with Gleason scores 6, 7, and ≥8 on biparametric magnetic resonance imaging (bpMRI) of the prostate using radiomics and to evaluate the added value of image augmentation and quantitative T1 mapping. MATERIALS AND METHODS: Eighty-five patients with subsequently histologically proven PCa underwent bpMRI at 3 T (T2-weighted imaging, diffusion-weighted imaging) with 66 patients undergoing additional T1 mapping at 3 T. The PCa lesions as well as the peripheral and transition zones were segmented pixel by pixel in multiple slices of the 3D MRI data sets (T2-weighted images, apparent diffusion coefficient, and T1 maps). To increase the size of the data set, images were augmented for contrast, brightness, noise, and perspective multiple times, effectively increasing the sample size 10-fold, and 322 different radiomics features were extracted before and after augmentation. Four different machine learning algorithms, including a random forest (RF), stochastic gradient boosting (SGB), support vector machine (SVM), and k-nearest neighbor, were trained with and without features from T1 maps to differentiate among 3 different Gleason groups (6, 7, and ≥8). RESULTS: Support vector machine showed the highest accuracy of 0.92 (95% confidence interval [CI], 0.62-1.00) for classifying the different Gleason scores, followed by RF (0.83; 95% CI, 0.52-0.98), SGB (0.75; 95% CI, 0.43-0.95), and k-nearest neighbor (0.50; 95% CI, 0.21-0.79). Image augmentation resulted in an average increase in accuracy between 0.08 (SGB) and 0.48 (SVM). Removing T1 mapping features led to a decline in accuracy for RF (-0.16) and SGB (-0.25) and a higher generalization error. CONCLUSIONS: When data are limited, image augmentations and features from quantitative T1 mapping sequences might help to achieve higher accuracy and lower generalization error for classification among different Gleason groups in bpMRI by using
Chen C, Hammernik K, Ouyang C, et al., 2021, Cooperative training and latent space data augmentation for robust medical image segmentation, International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)
Wang S, Qin C, Savioli N, et al., 2021, Joint motion correction and super resolution for cardiac segmentationvia latent optimisation, International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), Publisher: Springer, Pages: 14-24
In cardiac magnetic resonance (CMR) imaging, a 3D high-resolution segmentation of the heart is essential for detailed description of its anatomical structures. However, due to the limit of acquisition duration andrespiratory/cardiac motion, stacks of multi-slice 2D images are acquired inclinical routine. The segmentation of these images provides a low-resolution representation of cardiac anatomy, which may contain artefacts caused by motion. Here we propose a novel latent optimisation framework that jointly performs motion correction and super resolution for cardiac image segmentations. Given a low-resolution segmentation as input, the framework accounts for inter-slice motion in cardiac MR imaging and super-resolves the input into a high-resolution segmentation consistent with input. A multi-view loss is incorporated to leverage information from both short-axis view and long-axis view of cardiac imaging. To solve the inverse problem, iterative optimisation is performed in a latent space, which ensures the anatomical plausibility. This alleviates the need of paired low-resolution and high-resolution images for supervised learning. Experiments on two cardiac MR datasets show that the proposed framework achieves high performance, comparable to state-of-the-art super-resolution approaches and with better cross-domain generalisability and anatomical plausibility.
Hammernik K, Schlemper J, Qin C, et al., 2021, Systematic evaluation of iterative deep neural networks for fast parallel MRI reconstruction with sensitivity-weighted coil combination, MAGNETIC RESONANCE IN MEDICINE, Vol: 86, Pages: 1859-1872, ISSN: 0740-3194
- Author Web Link
- Cite
- Citations: 24
van Veen E, van der Jagt M, Citerio G, et al., 2021, Occurrence and timing of withdrawal of life-sustaining measures in traumatic brain injury patients: a CENTER-TBI study., Intensive Care Med, Vol: 47, Pages: 1115-1129
BACKGROUND: In patients with severe brain injury, withdrawal of life-sustaining measures (WLSM) is common in intensive care units (ICU). WLSM constitutes a dilemma: instituting WLSM too early could result in death despite the possibility of an acceptable functional outcome, whereas delaying WLSM could unnecessarily burden patients, families, clinicians, and hospital resources. We aimed to describe the occurrence and timing of WLSM, and factors associated with timing of WLSM in European ICUs in patients with traumatic brain injury (TBI). METHODS: The CENTER-TBI Study is a prospective multi-center cohort study. For the current study, patients with traumatic brain injury (TBI) admitted to the ICU and aged 16 or older were included. Occurrence and timing of WLSM were documented. For the analyses, we dichotomized timing of WLSM in early (< 72 h after injury) versus later (≥ 72 h after injury) based on recent guideline recommendations. We assessed factors associated with initiating WLSM early versus later, including geographic region, center, patient, injury, and treatment characteristics with univariable and multivariable (mixed effects) logistic regression. RESULTS: A total of 2022 patients aged 16 or older were admitted to the ICU. ICU mortality was 13% (n = 267). Of these, 229 (86%) patients died after WLSM, and were included in the analyses. The occurrence of WLSM varied between regions ranging from 0% in Eastern Europe to 96% in Northern Europe. In 51% of the patients, WLSM was early. Patients in the early WLSM group had a lower maximum therapy intensity level (TIL) score than patients in the later WLSM group (median of 5 versus 10) The strongest independent variables associated with early WLSM were one unreactive pupil (odds ratio (OR) 4.0, 95% confidence interval (CI) 1.3-12.4) or two unreactive pupils (OR 5.8, CI 2.6-13.1) compared to two reactive pupils, and an Injury Severity Score (ISS) if over 41 (OR per point above
Li L, Sinclair M, Makropoulos A, et al., 2021, CAS-Net: Conditional atlas generation and brain segmentation for fetal MRI, 24th International Conference on Medical Image Computing and Computer Assisted Intervention, Publisher: Springer, Pages: 221-230, ISSN: 0302-9743
Fetal Magnetic Resonance Imaging (MRI) is used in prenatal diagnosis and to assess early brain development. Accurate segmentation of the different brain tissues is a vital step in several brain analysis tasks, such as cortical surface reconstruction and tissue thickness measurements. Fetal MRI scans, however, are prone to motion artifacts that can affect the correctness of both manual and automatic segmentation techniques. In this paper, we propose a novel network structure that can simultaneously generate conditional atlases and predict brain tissue segmentation, called CAS-Net. The conditional atlases provide anatomical priors that can constrain the segmentation connectivity, despite the heterogeneity of intensity values caused by motion or partial volume effects. The proposed method is trained and evaluated on 253 subjects from the developing Human Connectome Project (dHCP). The results demonstrate that the proposed method can generate conditional age-specific atlas with sharp boundary and shape variance. It also segment multi-category brain tissues for fetal MRI with a high overall Dice similarity coefficient (DSC) of 85.2% for the selected 9 tissue labels.
Budd S, Sinclair M, Day T, et al., 2021, Detecting hypo-plastic left heart syndrome in fetal ultrasound via disease-specific atlas maps, 24th International Conference on Medical Image Computing and Computer Assisted Intervention, Publisher: Springer, Pages: 207-217, ISSN: 0302-9743
Fetal ultrasound screening during pregnancy plays a vital role in the early detection of fetal malformations which have potential long-term health impacts. The level of skill required to diagnose such malformations from live ultrasound during examination is high and resources for screening are often limited. We present an interpretable, atlas-learning segmentation method for automatic diagnosis of Hypo-plastic Left Heart Syndrome (HLHS) from a single ‘4 Chamber Heart’ view image. We propose to extend the recently introduced Image-and-Spatial Transformer Networks (Atlas-ISTN) into a framework that enables sensitising atlas generation to disease. In this framework we can jointly learn image segmentation, registration, atlas construction and disease prediction while providing a maximum level of clinical interpretability compared to direct image classification methods. As a result our segmentation allows diagnoses competitive with expert-derived manual diagnosis and yields an AUC-ROC of 0.978 (1043 cases for training, 260 for validation and 325 for testing).
Kamnitsas K, Winzeck S, Kornaropoulos EN, et al., 2021, Transductive image segmentation: Self-training and effect of uncertainty estimation, MICCAI Workshop on Domain Adaptation and Representation Transfer, Publisher: Springer, Pages: 79-89
Semi-supervised learning (SSL) uses unlabeled data during training to learnbetter models. Previous studies on SSL for medical image segmentation focusedmostly on improving model generalization to unseen data. In some applications,however, our primary interest is not generalization but to obtain optimalpredictions on a specific unlabeled database that is fully available duringmodel development. Examples include population studies for extracting imagingphenotypes. This work investigates an often overlooked aspect of SSL,transduction. It focuses on the quality of predictions made on the unlabeleddata of interest when they are included for optimization during training,rather than improving generalization. We focus on the self-training frameworkand explore its potential for transduction. We analyze it through the lens ofInformation Gain and reveal that learning benefits from the use of calibratedor under-confident models. Our extensive experiments on a large MRI databasefor multi-class segmentation of traumatic brain lesions shows promising resultswhen comparing transductive with inductive predictions. We believe this studywill inspire further research on transductive learning, a well-suited paradigmfor medical image analysis.
Ma Q, Robinson EC, Kainz B, et al., 2021, PialNN: A fast deep learning framework for cortical pial surface reconstruction, 24th International Conference on Medical Image Computing and Computer Assisted Intervention, Publisher: Springer, Pages: 73-81, ISSN: 0302-9743
Traditional cortical surface reconstruction is time consuming and limited by the resolution of brain Magnetic Resonance Imaging (MRI). In this work, we introduce Pial Neural Network (PialNN), a 3D deep learning framework for pial surface reconstruction. PialNN is trained end-to-end to deform an initial white matter surface to a target pial surface by a sequence of learned deformation blocks. A local convolutional operation is incorporated in each block to capture the multi-scale MRI information of each vertex and its neighborhood. This is fast and memory-efficient, which allows reconstructing a pial surface mesh with 150k vertices in one second. The performance is evaluated on the Human Connectome Project (HCP) dataset including T1-weighted MRI scans of 300 subjects. The experimental results demonstrate that PialNN reduces the geometric error of the predicted pial surface by 30% compared to state-of-the-art deep learning approaches. The codes are publicly available at https://github.com/m-qiang/PialNN.
Tan J, Hou B, Day T, et al., 2021, Detecting outliers with poisson image interpolation, 24th International Conference on Medical Image Computing and Computer Assisted Intervention, Publisher: Springer, Pages: 581-591, ISSN: 0302-9743
Supervised learning of every possible pathology is unrealistic for many primary care applications like health screening. Image anomaly detection methods that learn normal appearance from only healthy data have shown promising results recently. We propose an alternative to image reconstruction-based and image embedding-based methods and propose a new self-supervised method to tackle pathological anomaly detection. Our approach originates in the foreign patch interpolation (FPI) strategy that has shown superior performance on brain MRI and abdominal CT data. We propose to use a better patch interpolation strategy, Poisson image interpolation (PII), which makes our method suitable for applications in challenging data regimes. PII outperforms state-of-the-art methods by a good margin when tested on surrogate tasks like identifying common lung anomalies in chest X-rays or hypo-plastic left heart syndrome in prenatal, fetal cardiac ultrasound images. Code available at https://github.com/jemtan/PII.
Simoes Monteiro de Marvao A, McGurk K, Zheng S, et al., 2021, Phenotypic expression and outcomes in individuals with rare genetic variants of hypertrophic cardiomyopathy, Journal of the American College of Cardiology, Vol: 78, Pages: 1097-1110, ISSN: 0735-1097
Background: Hypertrophic cardiomyopathy (HCM) is caused by rare variants in sarcomereencoding genes, but little is known about the clinical significance of these variants in thegeneral population.Objectives: To compare lifetime outcomes and cardiovascular phenotypes according to thepresence of rare variants in sarcomere-encoding genes amongst middle-aged adults.Methods: We analysed whole exome sequencing and cardiac magnetic resonance (CMR)imaging in UK Biobank participants stratified by sarcomere-encoding variant status.Results: The prevalence of rare variants (allele frequency <0.00004) in HCM-associatedsarcomere-encoding genes in 200,584 participants was 2.9% (n=5,712; 1 in 35), and theprevalence of variants pathogenic or likely pathogenic for HCM (SARC-HCM-P/LP) was0.25% (n=493, 1 in 407). SARC-HCM-P/LP variants were associated with increased risk ofdeath or major adverse cardiac events compared to controls (HR 1.69, 95% CI 1.38 to 2.07,p<0.001), mainly due to heart failure endpoints (HR 4.23, 95% CI 3.07 to 5.83, p<0.001). In21,322 participants with CMR, SARC-HCM-P/LP were associated with asymmetric increasein left ventricular maximum wall thickness (10.9±2.7 vs 9.4±1.6 mm, p<0.001) buthypertrophy (≥13mm) was only present in 18.4% (n=9/49, 95% CI 9 to 32%). SARC-HCMP/LP were still associated with heart failure after adjustment for wall thickness (HR 6.74,95% CI 2.43 to 18.7, p<0.001).Conclusions: In this population of middle-aged adults, SARC-HCM-P/LP variants have lowaggregate penetrance for overt HCM but are associated with increased risk of adversecardiovascular outcomes and an attenuated cardiomyopathic phenotype. Although absoluteevent rates are low, identification of these variants may enhance risk stratification beyondfamilial disease.
Lecky FE, Otesile O, Marincowitz C, et al., 2021, The burden of traumatic brain injury from low-energy falls among patients from 18 countries in the CENTER-TBI Registry: A comparative cohort study., PLoS Med, Vol: 18
BACKGROUND: Traumatic brain injury (TBI) is an important global public health burden, where those injured by high-energy transfer (e.g., road traffic collisions) are assumed to have more severe injury and are prioritised by emergency medical service trauma triage tools. However recent studies suggest an increasing TBI disease burden in older people injured through low-energy falls. We aimed to assess the prevalence of low-energy falls among patients presenting to hospital with TBI, and to compare their characteristics, care pathways, and outcomes to TBI caused by high-energy trauma. METHODS AND FINDINGS: We conducted a comparative cohort study utilising the CENTER-TBI (Collaborative European NeuroTrauma Effectiveness Research in TBI) Registry, which recorded patient demographics, injury, care pathway, and acute care outcome data in 56 acute trauma receiving hospitals across 18 countries (17 countries in Europe and Israel). Patients presenting with TBI and indications for computed tomography (CT) brain scan between 2014 to 2018 were purposively sampled. The main study outcomes were (i) the prevalence of low-energy falls causing TBI within the overall cohort and (ii) comparisons of TBI patients injured by low-energy falls to TBI patients injured by high-energy transfer-in terms of demographic and injury characteristics, care pathways, and hospital mortality. In total, 22,782 eligible patients were enrolled, and study outcomes were analysed for 21,681 TBI patients with known injury mechanism; 40% (95% CI 39% to 41%) (8,622/21,681) of patients with TBI were injured by low-energy falls. Compared to 13,059 patients injured by high-energy transfer (HE cohort), the those injured through low-energy falls (LE cohort) were older (LE cohort, median 74 [IQR 56 to 84] years, versus HE cohort, median 42 [IQR 25 to 60] years; p < 0.001), more often female (LE cohort, 50% [95% CI 48% to 51%], versus HE cohort, 32% [95% CI 31% to 34%]; p < 0.001), more frequently taking pre-inj
Usynin D, Ziller A, Makowski M, et al., 2021, Adversarial interference and its mitigations in privacy-preserving collaborative machine learning, Nature Machine Intelligence, Vol: 3, Pages: 749-758, ISSN: 2522-5839
Despite the rapid increase of data available to train machine-learning algorithms in many domains, several applications suffer from a paucity of representative and diverse data. The medical and financial sectors are, for example, constrained by legal, ethical, regulatory and privacy concerns preventing data sharing between institutions. Collaborative learning systems, such as federated learning, are designed to circumvent such restrictions and provide a privacy-preserving alternative by eschewing data sharing and relying instead on the distributed remote execution of algorithms. However, such systems are susceptible to malicious adversarial interference attempting to undermine their utility or divulge confidential information. Here we present an overview and analysis of current adversarial attacks and their mitigations in the context of collaborative machine learning. We discuss the applicability of attack vectors to specific learning contexts and attempt to formulate a generic foundation for adversarial influence and mitigation mechanisms. We moreover show that a number of context-specific learning conditions are exploited in similar fashion across all settings. Lastly, we provide a focused perspective on open challenges and promising areas of future research in the field.
Sewalt CA, Gravesteijn BY, Menon D, et al., 2021, Primary versus early secondary referral to a specialized neurotrauma center in patients with moderate/severe traumatic brain injury: a CENTER TBI study., Scand J Trauma Resusc Emerg Med, Vol: 29
BACKGROUND: Prehospital care for patients with traumatic brain injury (TBI) varies with some emergency medical systems recommending direct transport of patients with moderate to severe TBI to hospitals with specialist neurotrauma care (SNCs). The aim of this study is to assess variation in levels of early secondary referral within European SNCs and to compare the outcomes of directly admitted and secondarily transferred patients. METHODS: Patients with moderate and severe TBI (Glasgow Coma Scale < 13) from the prospective European CENTER-TBI study were included in this study. All participating hospitals were specialist neuroscience centers. First, adjusted between-country differences were analysed using random effects logistic regression where early secondary referral was the dependent variable, and a random intercept for country was included. Second, the adjusted effect of early secondary referral on survival to hospital discharge and functional outcome [6 months Glasgow Outcome Scale Extended (GOSE)] was estimated using logistic and ordinal mixed effects models, respectively. RESULTS: A total of 1347 moderate/severe TBI patients from 53 SNCs in 18 European countries were included. Of these 1347 patients, 195 (14.5%) were admitted after early secondary referral. Secondarily referred moderate/severe TBI patients presented more often with a CT abnormality: mass lesion (52% vs. 34%), midline shift (54% vs. 36%) and acute subdural hematoma (77% vs. 65%). After adjusting for case-mix, there was a large European variation in early secondary referral, with a median OR of 1.69 between countries. Early secondary referral was not associated with functional outcome (adjusted OR 1.07, 95% CI 0.78-1.69), nor with survival at discharge (1.05, 0.58-1.90). CONCLUSIONS: Across Europe, substantial practice variation exists in the proportion of secondarily referred TBI patients at SNCs that is not explained by case mix. Within SNCs early secondary referral does
Böhm JK, Güting H, Thorn S, et al., 2021, Global Characterisation of Coagulopathy in Isolated Traumatic Brain Injury (iTBI): A CENTER-TBI Analysis., Neurocrit Care, Vol: 35, Pages: 184-196
BACKGROUND: Trauma-induced coagulopathy in patients with traumatic brain injury (TBI) is associated with high rates of complications, unfavourable outcomes and mortality. The mechanism of the development of TBI-associated coagulopathy is poorly understood. METHODS: This analysis, embedded in the prospective, multi-centred, observational Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) study, aimed to characterise the coagulopathy of TBI. Emphasis was placed on the acute phase following TBI, primary on subgroups of patients with abnormal coagulation profile within 4 h of admission, and the impact of pre-injury anticoagulant and/or antiplatelet therapy. In order to minimise confounding factors, patients with isolated TBI (iTBI) (n = 598) were selected for this analysis. RESULTS: Haemostatic disorders were observed in approximately 20% of iTBI patients. In a subgroup analysis, patients with pre-injury anticoagulant and/or antiplatelet therapy had a twice exacerbated coagulation profile as likely as those without premedication. This was in turn associated with increased rates of mortality and unfavourable outcome post-injury. A multivariate analysis of iTBI patients without pre-injury anticoagulant therapy identified several independent risk factors for coagulopathy which were present at hospital admission. Glasgow Coma Scale (GCS) less than or equal to 8, base excess (BE) less than or equal to - 6, hypothermia and hypotension increased risk significantly. CONCLUSION: Consideration of these factors enables early prediction and risk stratification of acute coagulopathy after TBI, thus guiding clinical management.
Wiegers EJA, Lingsma HF, Huijben JA, et al., 2021, Fluid balance and outcome in critically ill patients with traumatic brain injury (CENTER-TBI and OzENTER-TBI): a prospective, multicentre, comparative effectiveness study., Lancet Neurol, Vol: 20, Pages: 627-638
BACKGROUND: Fluid therapy-the administration of fluids to maintain adequate organ tissue perfusion and oxygenation-is essential in patients admitted to the intensive care unit (ICU) with traumatic brain injury. We aimed to quantify the variability in fluid management policies in patients with traumatic brain injury and to study the effect of this variability on patients' outcomes. METHODS: We did a prospective, multicentre, comparative effectiveness study of two observational cohorts: CENTER-TBI in Europe and OzENTER-TBI in Australia. Patients from 55 hospitals in 18 countries, aged 16 years or older with traumatic brain injury requiring a head CT, and admitted to the ICU were included in this analysis. We extracted data on demographics, injury, and clinical and treatment characteristics, and calculated the mean daily fluid balance (difference between fluid input and loss) and mean daily fluid input during ICU stay per patient. We analysed the association of fluid balance and input with ICU mortality and functional outcome at 6 months, measured by the Glasgow Outcome Scale Extended (GOSE). Patient-level analyses relied on adjustment for key characteristics per patient, whereas centre-level analyses used the centre as the instrumental variable. FINDINGS: 2125 patients enrolled in CENTER-TBI and OzENTER-TBI between Dec 19, 2014, and Dec 17, 2017, were eligible for inclusion in this analysis. The median age was 50 years (IQR 31 to 66) and 1566 (74%) of patients were male. The median of the mean daily fluid input ranged from 1·48 L (IQR 1·12 to 2·09) to 4·23 L (3·78 to 4·94) across centres. The median of the mean daily fluid balance ranged from -0·85 L (IQR -1·51 to -0·49) to 1·13 L (0·99 to 1·37) across centres. In patient-level analyses, a mean positive daily fluid balance was associated with higher ICU mortality (odds ratio [OR] 1·10 [95% CI 1·07 to 1·12] per 0&mi
Carney O, Hughes E, Tusor N, et al., 2021, Incidental findings on brain MR imaging of asymptomatic term neonates in the Developing Human Connectome Project, ECLINICALMEDICINE, Vol: 38
- Author Web Link
- Cite
- Citations: 10
Chai S, Rueckert D, Fetit A, 2021, Reducing textural bias improves robustness of deep segmentation models, Annual Conference on Medical Image Understanding and Analysis (MIUA 2021), Publisher: Springer Verlag, Pages: 294-304, ISSN: 0302-9743
Despite advances in deep learning, robustness under domain shift remains a major bottleneck in medical imaging settings. Findings on natural images suggest that deep neural models can show a strong textural bias when carrying out image classification tasks. In this thorough empirical study, we draw inspiration from findings on natural images and investigate ways in which addressing the textural bias phenomenon could bring up the robustness of deep segmentation models when applied to three-dimensional (3D) medical data. To achieve this, publicly available MRI scans from the Developing Human Connectome Project are used to study ways in which simulating textural noise can help train robust models in a complex semantic segmentation task. We contribute an extensive empirical investigation consisting of 176 experiments and illustrate how applying specific types of simulated textural noise prior to training can lead to texture invariant models, resulting in improved robustness when segmenting scans corrupted by previously unseen noise types and levels.
Eyre M, Fitzgibbon SP, Ciarrusta J, et al., 2021, The Developing Human Connectome Project: typical and disrupted perinatal functional connectivity, Brain: a journal of neurology, Vol: 144, Pages: 2199-2213, ISSN: 0006-8950
The Developing Human Connectome Project (dHCP) is an Open Science project which provides the first large sample of neonatal functional MRI (fMRI) data with high temporal and spatial resolution. This data enables mapping of intrinsic functional connectivity between spatially distributed brain regions under normal and adverse perinatal circumstances, offering a framework to study the ontogeny of large-scale brain organisation in humans. Here, we characterise in unprecedented detail the maturation and integrity of resting-state networks (RSNs) at term-equivalent age in 337 infants (including 65 born preterm). First, we applied group independent component analysis (ICA) to define 11 RSNs in term-born infants scanned at 43.5-44.5 weeks postmenstrual age (PMA). Adult-like topography was observed in RSNs encompassing primary sensorimotor, visual and auditory cortices. Among six higher-order, association RSNs, analogues of the adult networks for language and ocular control were identified, but a complete default mode network precursor was not. Next, we regressed the subject-level datasets from an independent cohort of infants scanned at 37-43.5 weeks PMA against the group-level RSNs to test for the effects of age, sex and preterm birth. Brain mapping in term-born infants revealed areas of positive association with age across four of six association RSNs, indicating active maturation in functional connectivity from 37 to 43.5 weeks PMA. Female infants showed increased connectivity in inferotemporal regions of the visual association network. Preterm birth was associated with striking impairments of functional connectivity across all RSNs in a dose-dependent manner; conversely, connectivity of the superior parietal lobules within the lateral motor network was abnormally increased in preterm infants, suggesting a possible mechanism for specific difficulties such as developmental coordination disorder which occur frequently in preterm children. Overall, we find a robust, modular
Ziller A, Usynin D, Braren R, et al., 2021, Medical imaging deep learning with differential privacy, Scientific Reports, Vol: 11, ISSN: 2045-2322
The successful training of deep learning models for diagnostic deployment in medical imaging applications requires large volumes of data. Such data cannot be procured without consideration for patient privacy, mandated both by legal regulations and ethical requirements of the medical profession. Differential privacy (DP) enables the provision of information-theoretic privacy guarantees to patients and can be implemented in the setting of deep neural network training through the differentially private stochastic gradient descent (DP-SGD) algorithm. We here present deepee, a free-and-open-source framework for differentially private deep learning for use with the PyTorch deep learning framework. Our framework is based on parallelised execution of neural network operations to obtain and modify the per-sample gradients. The process is efficiently abstracted via a data structure maintaining shared memory references to neural network weights to maintain memory efficiency. We furthermore offer specialised data loading procedures and privacy budget accounting based on the Gaussian Differential Privacy framework, as well as automated modification of the user-supplied neural network architectures to ensure DP-conformity of its layers. We benchmark our framework's computational performance against other open-source DP frameworks and evaluate its application on the paediatric pneumonia dataset, an image classification task and on the Medical Segmentation Decathlon Liver dataset in the task of medical image segmentation. We find that neural network training with rigorous privacy guarantees is possible while maintaining acceptable classification performance and excellent segmentation performance. Our framework compares favourably to related work with respect to memory consumption and computational performance. Our work presents an open-source software framework for differentially private deep learning, which we demonstrate in medical imaging analysis tasks. It serves to further th
Thanaj M, Mielke J, McGurk KA, et al., 2021, Genetic and environmental determinants of diastolic heart function
<jats:title>ABSTRACT</jats:title><jats:p>Diastole is the sequence of physiological events that occur in the heart during ventricular filling and principally depends on myocardial relaxation and chamber stiffness. Abnormal diastolic function is related to many cardiovascular disease processes and is predictive of health outcomes, but its genetic architecture is largely unknown. Here, we use machine learning cardiac motion analysis to measure diastolic functional traits in 39,559 participants of UK Biobank and perform a genome-wide association study. We identified 9 significant, independent loci near genes that are associated with maintaining sarcomeric function under biomechanical stress and genes implicated in the development of cardiomyopathy. Age, sex and diabetes were independent predictors of diastolic function and we found a causal relationship between ventricular stiffness and heart failure. Our results provide novel insights into the genetic and environmental factors influencing diastolic function that are relevant for identifying causal relationships and tractable targets in heart failure.</jats:p>
Kart T, Fischer M, Kuestner T, et al., 2021, Deep Learning-Based Automated Abdominal Organ Segmentation in the UK Biobank and German National Cohort Magnetic Resonance Imaging Studies, INVESTIGATIVE RADIOLOGY, Vol: 56, Pages: 401-408, ISSN: 0020-9996
- Author Web Link
- Cite
- Citations: 16
Kaissis G, Ziller A, Passerat-Palmbach J, et al., 2021, End-to-end privacy preserving deep learning on multi-institutional medical imaging, Nature Machine Intelligence, Vol: 3, Pages: 473-484, ISSN: 2522-5839
Using large, multi-national datasets for high-performance medical imaging AI systems requires innovation in privacy-preserving machine learning so models can train on sensitive data without requiring data transfer. Here we present PriMIA (Privacy-preserving Medical Image Analysis), a free, open-source software framework for differentially private, securely aggregated federated learning and encrypted inference on medical imaging data. We test PriMIA using a real-life case study in which an expert-level deep convolutional neural network classifies paediatric chest X-rays; the resulting model’s classification performance is on par with locally, non-securely trained models. We theoretically and empirically evaluate our framework’s performance and privacy guarantees, and demonstrate that the protections provided prevent the reconstruction of usable data by a gradient-based model inversion attack. Finally, we successfully employ the trained model in an end-to-end encrypted remote inference scenario using secure multi-party computation to prevent the disclosure of the data and the model.
Zhou SK, Greenspan H, Davatzikos C, et al., 2021, A Review of Deep Learning in Medical Imaging: Imaging Traits, Technology Trends, Case Studies With Progress Highlights, and Future Promises, PROCEEDINGS OF THE IEEE, Vol: 109, Pages: 820-838, ISSN: 0018-9219
- Author Web Link
- Cite
- Citations: 171
Dimitrova R, Arulkumaran S, Carney O, et al., 2021, Phenotyping the preterm brain: characterizing individual deviations from normative volumetric development in two large infant cohorts, Cerebral Cortex, Vol: 31, Pages: 3665-3677, ISSN: 1047-3211
The diverse cerebral consequences of preterm birth create significant challenges for understanding pathogenesis or predicting later outcome. Instead of focusing on describing effects common to the group, comparing individual infants against robust normative data offers a powerful alternative to study brain maturation. Here we used Gaussian process regression to create normative curves characterizing brain volumetric development in 274 term-born infants, modeling for age at scan and sex. We then compared 89 preterm infants scanned at term-equivalent age with these normative charts, relating individual deviations from typical volumetric development to perinatal risk factors and later neurocognitive scores. To test generalizability, we used a second independent dataset comprising of 253 preterm infants scanned using different acquisition parameters and scanner. We describe rapid, nonuniform brain growth during the neonatal period. In both preterm cohorts, cerebral atypicalities were widespread, often multiple, and varied highly between individuals. Deviations from normative development were associated with respiratory support, nutrition, birth weight, and later neurocognition, demonstrating their clinical relevance. Group-level understanding of the preterm brain disguises a large degree of individual differences. We provide a method and normative dataset that offer a more precise characterization of the cerebral consequences of preterm birth by profiling the individual neonatal brain.
Jaubert O, Cruz G, Bustin A, et al., 2021, T1, T2, and Fat Fraction Cardiac MR Fingerprinting: Preliminary Clinical Evaluation, JOURNAL OF MAGNETIC RESONANCE IMAGING, Vol: 53, Pages: 1253-1265, ISSN: 1053-1807
- Author Web Link
- Cite
- Citations: 22
Dou Q, So TY, Jiang M, et al., 2021, Federated deep learning for detecting COVID-19 lung abnormalities in CT: a privacy-preserving multinational validation study, npj Digital Medicine, Vol: 4, Pages: 1-11, ISSN: 2398-6352
Data privacy mechanisms are essential for rapidly scaling medical training databases to capture the heterogeneity of patient data distributions toward robust and generalizable machine learning systems. In the current COVID-19 pandemic, a major focus of artificial intelligence (AI) is interpreting chest CT, which can be readily used in the assessment and management of the disease. This paper demonstrates the feasibility of a federated learning method for detecting COVID-19 related CT abnormalities with external validation on patients from a multinational study. We recruited 132 patients from seven multinational different centers, with three internal hospitals from Hong Kong for training and testing, and four external, independent datasets from Mainland China and Germany, for validating model generalizability. We also conducted case studies on longitudinal scans for automated estimation of lesion burden for hospitalized COVID-19 patients. We explore the federated learning algorithms to develop a privacy-preserving AI model for COVID-19 medical image diagnosis with good generalization capability on unseen multinational datasets. Federated learning could provide an effective mechanism during pandemics to rapidly develop clinically useful AI across institutions and countries overcoming the burden of central aggregation of large amounts of sensitive data.
Meng Q, Matthew J, Zimmer VA, et al., 2021, Mutual information-based disentangled neural networks for classifying unseen categories in different domains: application to fetal ultrasound imaging, IEEE Transactions on Medical Imaging, Vol: 40, Pages: 722-734, ISSN: 0278-0062
Deep neural networks exhibit limited generalizability across images with different entangled domain features and categorical features. Learning generalizable features that can form universal categorical decision boundaries across domains is an interesting and difficult challenge. This problem occurs frequently in medical imaging applications when attempts are made to deploy and improve deep learning models across different image acquisition devices, across acquisition parameters or if some classes are unavailable in new training databases. To ad-dress this problem, we propose Mutual Information-based Disentangled Neural Networks (MIDNet), which extract generalizable categorical features to transfer knowledge to unseen categories in a target domain. The proposed MID-Net adopts a semi-supervised learning paradigm to alleviate the dependency on labeled data. This is important for real-world applications where data annotation is time-consuming, costly and requires training and expertise. We extensively evaluate the proposed method on fetal ultra-sound datasets for two different image classification tasks where domain features are respectively defined by shadow artifacts and image acquisition devices. Experimental results show that the proposed method outperforms the state-of-the-art on the classification of unseen categories in a target domain with sparsely labeled training data.
Balaban G, Halliday B, Bradley P, et al., 2021, Late-gadolinium enhancement interface area and electrophysiological simulations predict arrhythmic events in non-ischemic dilated cardiomyopathy patients, JACC: Clinical Electrophysiology, Vol: 7, Pages: 238-249, ISSN: 2405-5018
BACKGROUND: The presence of late-gadolinium enhancement (LGE) predicts life threatening ventricular arrhythmias in non-ischemic dilated cardiomyopathy (NIDCM); however, risk stratification remains imprecise. LGE shape and simulations of electrical activity may be able to provide additional prognostic information.OBJECTIVE: This study sought to investigate whether shape-based LGE metrics and simulations of reentrant electrical activity are associated with arrhythmic events in NIDCM patients.METHODS: CMR-LGE shape metrics were computed for a cohort of 156 NIDCM patients with visible LGE and tested retrospectively for an association with an arrhythmic composite end-point of sudden cardiac death and ventricular tachycardia. Computational models were created from images and used in conjunction with simulated stimulation protocols to assess the potential for reentry induction in each patient’s scar morphology. A mechanistic analysis of the simulations was carried out to explain the associations. RESULTS: During a median follow-up of 1611 [IQR 881-2341] days, 16 patients (10.3%) met the primary endpoint. In an inverse probability weighted Cox regression, the LGE-myocardial interface area (HR:1.75; 95% CI:1.24-2.47; p=0.001), number of simulated reentries (HR: 1.4; 95% CI: 1.23-1.59; p<0.01) and LGE volume (HR:1.44; 95% CI:1.07-1.94; p=0.02) were associated with arrhythmic events. Computational modeling revealed repolarisation heterogeneity and rate-dependent block of electrical wavefronts at the LGE-myocardial interface as putative arrhythmogenic mechanisms directly related to LGE interface area.CONCLUSION: The area of interface between scar and surviving myocardium, as well as simulated reentrant activity, are associated with an elevated risk of major arrhythmic events in NIDCM patients with LGE and represent novel risk predictors.
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.