Search

Magnetic Resonance Fingerprinting (MRF) enables simultaneous mapping of multiple tissue parameters such as T1 and T2 relaxation times. The working principle of MRF relies on varying acquisition parameters pseudo-randomly, so that each tissue generates its unique signal evolution during scanning. Even though MRF provides faster scanning, it has disadvantages such as erroneous and slow generation of the corresponding parametric maps, which needs to be improved. Moreover, there is a need for explainable architectures for understanding the guiding signals to generate accurate parametric maps.In this paper, we addressed both of these shortcomings by proposing a novel neural network architecture (CONV-ICA) consisting of a channel-wise attention module and a fully convolutional network. Another contribution of this study is a new channel selection method: attention-based channel selection. Furthermore, the effect of patch size and temporal frames of MRF signal on channel reduction are analyzed by employing a channel-wise attention.The proposed approach, evaluated over 3 simulated MRF signals, reduces error in the reconstruction of tissue parameters by 8.88% for T1 and 75.44% for T2 with respect to state-of-the-art methods.It is demonstrated that channel attention mechanism helps to focus on informative channels and fully convolutional network extracts spatial information achieve the best reconstruction performance.As a consequence of improvement in fast and accurate manner, presented work can contribute to make MRF appropriate for clinical use.

ObjectiveDynamic magnetic resonance (MR) imaging enables visualisation of articulators during speech. There is growing interest in quantifying articulator motion in two-dimensional MR images of the vocal tract, to better understand speech production and potentially inform patient management decisions. Image registration is an established way to achieve this quantification. Recently, segmentation-informed deformable registration frameworks have been developed and have achieved state-of-the-art accuracy. This work aims to adapt such a framework and optimise it for estimating displacement fields between dynamic two-dimensional MR images of the vocal tract during speech.MethodsA deep-learning-based registration framework was developed and compared with current state-of-the-art registration methods and frameworks (two traditional methods and three deep-learning-based frameworks, two of which are segmentation informed). The accuracy of the methods and frameworks was evaluated using the Dice coefficient (DSC), average surface distance (ASD) and a metric based on velopharyngeal closure. The metric evaluated if the fields captured a clinically relevant and quantifiable aspect of articulator motion.ResultsThe segmentation-informed frameworks achieved higher DSCs and lower ASDs and captured more velopharyngeal closures than the traditional methods and the framework that was not segmentation informed. All segmentation-informed frameworks achieved similar DSCs and ASDs. However, the proposed framework captured the most velopharyngeal closures.ConclusionsA framework was successfully developed and found to more accurately estimate articulator motion than five current state-of-the-art methods and frameworks.SignificanceThe first deep-learning-based framework specifically for registering dynamic two-dimensional MR images of the vocal tract during speech has been developed and evaluated.

AimsExisting strategies that identify post-infarct ventricular tachycardia (VT) ablation target either employ invasive electrophysiological (EP) mapping or non-invasive modalities utilizing the electrocardiogram (ECG). Their success relies on localizing sites critical to the maintenance of the clinical arrhythmia, not always recorded on the 12-lead ECG. Targeting the clinical VT by utilizing electrograms (EGM) recordings stored in implanted devices may aid ablation planning, enhancing safety and speed and potentially reducing the need of VT induction. In this context, we aim to develop a non-invasive computational-deep learning (DL) platform to localize VT exit sites from surface ECGs and implanted device intracardiac EGMs.Methods and resultsA library of ECGs and EGMs from simulated paced beats and representative post-infarct VTs was generated across five torso models. Traces were used to train DL algorithms to localize VT sites of earliest systolic activation; first tested on simulated data and then on a clinically induced VT to show applicability of our platform in clinical settings. Localization performance was estimated via localization errors (LEs) against known VT exit sites from simulations or clinical ablation targets. Surface ECGs successfully localized post-infarct VTs from simulated data with mean LE = 9.61 ± 2.61 mm across torsos. VT localization was successfully achieved from implanted device intracardiac EGMs with mean LE = 13.10 ± 2.36 mm. Finally, the clinically induced VT localization was in agreement with the clinical ablation volume.ConclusionThe proposed framework may be utilized for direct localization of post-infarct VTs from surface ECGs and/or implanted device EGMs, or in conjunction with efficient, patient-specific modelling, enhancing safety and speed of ablation planning.

Manifold alignment (MA) is a technique to map many high-dimensional datasets to one shared low-dimensional space. Here we develop a pipeline for using MA to reconstruct high-resolution medical images. We present two key contributions. Firstly, we develop a novel MA scheme in which each high-dimensional dataset can be differently weighted preventing noisier or less informative data from corrupting the aligned embedding. We find that this generalisation improves performance in our experiments in both supervised and unsupervised MA problems. Secondly, we use the wave kernel signature as a graph descriptor for the unsupervised MA case finding that it significantly outperforms the current state-of-the-art methods and provides higher quality reconstructed magnetic resonance volumes than existing methods.

Purpose: To develop a knowledge-based decision-support system capable of stratifying patients for rectal spacer (RS) insertion based on neural network predicted rectal dose, reducing the need for time- and resource-intensive radiotherapy (RT) planning. Methods: Forty-four patients treated for prostate cancer were enrolled into a clinical trial (NCT03238170). Dose-escalated prostate RT plans were manually created for 30 patients with simulated boost volumes using a conventional treatment planning system (TPS) and used to train a hierarchically dense 3D convolutional neural network to rapidly predict RT dose distributions. The network was used to predict rectal doses for 14 unseen test patients, with associated toxicity risks calculated according to published data. All metrics obtained using the network were compared to conventionally planned values. Results: The neural network stratified patients with an accuracy of 100% based on optimal rectal dose–volume histogram constraints and 78.6% based on mandatory constraints. The network predicted dose-derived grade 2 rectal bleeding risk within 95% confidence limits of -1.9% to +1.7% of conventional risk estimates (risk range 3.5%–9.9%) and late grade 2 fecal incontinence risk within -0.8% to +1.5% (risk range 2.3%–5.7%). Prediction of high-resolution 3D dose distributions took 0.7 s. Conclusions: The feasibility of using a neural network to provide rapid decision support for RS insertion prior to RT has been demonstrated, and the potential for time and resource savings highlighted. Directly after target and healthy tissue delineation, the network is able to (i) risk stratify most patients with a high degree of accuracy to prioritize which patients would likely derive greatest benefit from RS insertion and (ii) identify patients close to the stratification threshold who would require conventional planning.

Congenital heart disease (CHD) is the most commonly diagnosed birth defect. T2w black blood MRI provides optimal vessel visualisation, aiding prenatal CHD diagnosis. Common clinical practice involves manual segmentation of fetal heart and vessels for visualisation and reporting purposes.We propose an automated multi-label fetal cardiac vessels deep learning segmentation approach for T2w black blood MRI. Our network is trained using single-label manual segmentations obtained through current clinical practice, combined with a multi-label anatomical atlas with desired multi-label segmentation protocol. Our framework combines deep learning label propagation with 3D residual U-Net segmentation to produce high-quality multi-label output well adapted to the individual subject anatomy.We train and evaluate the network using forty fetal subjects with suspected coarctation of the aorta, achieving a dice score of 0.79 ± 0.02 for the fetal cardiac vessels region. The proposed network outperforms the label propagation and achieves a statistically equivalent performance to a 3D residual U-Net trained exclusively on manual single-label data (p-value>0.05). This multi-label framework therefore represents an advancement over the single-label approach, providing label-specific anatomical information, particularly useful for assessing specific anomaly areas in CHD.

Simpson's biplane rule (SBR) is considered the gold standard method for left ventricle (LV) volume quantification from echocardiography but relies on a summation-of-disks approach that makes assumptions about LV orientation and cross-sectional shape. We aim to identify key limiting factors in SBR and to develop a new robust standard for volume quantification. Three methods for computing LV volume were studied: (i) SBR, (ii) addition of a truncated basal cone (TBC) to SBR and (iii) a novel method of basal-oriented disks (BODs). Three retrospective cohorts representative of the young, adult healthy and heart failure populations were used to study the impact of anatomical variations in volume computations. Results reveal how basal slanting can cause over- and underestimation of volume, with errors by SBR and TBC10 mL for slanting angles6°. Only the BOD method correctly accounted for basal slanting, reducing relative volume errors by SBR from -2.23 ± 2.21% to -0.70 ± 1.91% in the adult population and similar qualitative performance in the other two cohorts. In conclusion, the summation of basal oriented disks, a novel interpretation of SBR, is a more accurate and precise method for estimating LV volume.

Background/Introduction Artificial intelligence (AI) is providing opportunities to transform cardiovascular medicine. A particular challenge in the application of AI technology is their potential for intrinsic and extrinsic biases, such as those based on gender and/or ethnicity. Unless satisfactorily addressed, these biases could lead to inequalities in early diagnosis, treatments and outcomes. Fairness in AI is a relatively new but fast-growing research field which deals with assessing and addressing potential bias in AI models. Purpose To perform the first analysis that assesses bias in AI-based cardiac MR segmentation models in a large-scale database. Methods A state-of-the-art deep learning (DL) based segmentation network, the “nnU-Net” framework [1], was used for automatic segmentation of both ventricles and the myocardium from cine short-axis cardiac MR over the full cardiac cycle. The dataset used consisted of end-diastole and end-systole short-axis cine cardiac MR images of 5,903 subjects (61.5±7.1 years). The nnU-Net network was trained and evaluated using a 5-fold cross validation (splits: train 60% / validation 20% / test 20%). Data on race and gender were obtained from the UK Biobank database and their distribution is summarized in Figure 1. To assess gender and racial bias in the segmentation network, we compared the Dice scores - which measure the overlap between manual and automatic segmentations – and the absolute error in measurements of biventricular volumes and function between patients grouped by ethnicity and gender. Results Figure 2 shows the Dice scores and the volumetric and functional measures for the full database, stratified by gender and by ethnicity. Results on the overall population showed an excellent agreement between the manual and automatic segmentations which is consistent with previous reported results [2–3]. However, we find statistically significant differences in Dice scores as well as volumetric measures between different ethnicities, showing that the segmentation network is biased against minority racial groups. No significant differences were found in Dice scores between genders. Similarly, for the end diastolic, end systolic volumes and ejection fraction, there were statistically significant differences in absolute error between the overall population and all racial groups except white. Conclusion(s) We have shown, for the first time, that racial bias exists in DL-based cardiac MR segmentation models. Our hypothesis is that this bias is a result of the unbalanced nature of the training data, and this is supported by the results which show that there is racial bias but not gender bias when trained using the UK Biobank database, which is gender-balanced but not race-balanced. In this work we want to highlight the potential issue of bias in DL-based image segmentation models when translating into a clinical environment. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): - EPSRC- Wellcome EPSRC Centre for Medical Engineering at the School of Biomedical Engineering and Imaging Sciences, King's College London Figure 1Figure 2

Background/Introduction Pressure-volume loops (PVloops) provide a wealth of information on cardiac function that is not readily available from cardiac imaging alone. Methods To estimate left ventricular (LV) PVloops non-invasively have been available, but have so far not been used to interrogate ventricular function in large patient cohorts, due to the complexity of estimating PVloops. A new method was recently validated that construct PVloops non-invasively from cine cardiac magnetic resonance (CMR), based on the time-varying elastance model [1]. At the same time, we have validated a framework for automated, quality controlled analysis of cine CMR in large cohorts of patients/subjects [2]. Combining these two methods could automated PVloop estimation, enabling analysis of ventricular pressure-volume relationships in large study populations. Purpose Evaluate if CMR-based non-invasive PVloops can be used to interrogate the impact of cardiac ageing on LV function occurring in a large population of healthy community dwellers. Methods Non-invasive PVloops were calculated from a full cardiac cycle LV volume curve and brachial blood pressure data using a recently validated method based on the time-varying elastance model [1], in 7,650 healthy community dwellers from the UKBiobank population study. The LV volume curve was automatically obtained using our state-of-the-art, quality controlled deep learning (DL) based cine CMR analysis framework [2]. External Work, pressure-volume-area (PVA), end-systolic pressure (Pes), ventricular elastance (Ees, an estimate of contractility) and arterial elastance (Ea) and energy per ejected volume (EEV: PVA/ stroke volume) were calculated from the PVloops. We performed univariate regression between PVloop parameters and age. We also calculated the additional impact of cardiovascular risk-factors in a multivariate analysis. Results See results in table 1. With age, LV volumes fall (p Conclusion Non-invasive CMR-based PVloop analyses capture the impact of known changes occurring during cardiac ageing on cardiac work, contractility and energetic expenditure. Obtaining PVloops automatically using our AI analysis system in this large cohort of healthy subjects allows to formulate reference for assessment of cardiac disease. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): The authors acknowledge financial support (support) the National Institute for Health Research (NIHR) Cardiovascular MedTech Co-operative (previously existing as the Cardiovascular Healthcare Technology Co-operative 2012 - 2017) award to the Guy's and St Thomas' NHS Foundation Trust, in partnership with King's College London and the NIHR comprehensive Biomedical Research Centre of the Guy's & St Thomas' NHS Foundation Trust. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health Univariate regression analysisExample of estimated PV loop

BackgroundArtificial intelligence (AI) techniques have been proposed for automation of cine CMR segmentation for functional quantification. However, in other applications AI models have been shown to have potential for sex and/or racial bias.ObjectivesTo perform the first analysis of sex/racial bias in AI-based cine CMR segmentation using a large-scale database.MethodsA state-of-the-art deep learning (DL) model was used for automatic segmentation of both ventricles and the myocardium from cine short-axis CMR. The dataset consisted of end-diastole and end-systole short-axis cine CMR images of 5,903 subjects from the UK Biobank database (61.5±7.1 years, 52% male, 81% white). To assess sex and racial bias, we compared Dice scores and errors in measurements of biventricular volumes and function between patients grouped by race and sex. To investigate whether segmentation bias could be explained by potential confounders, a multivariate linear regression and ANCOVA were performed.ResultsWe found statistically significant differences in Dice scores (white ∼94% vs minority ethnic groups 86-89%) as well as in absolute/relative errors in volumetric and functional measures, showing that the AI model was biased against minority racial groups, even after correction for possible confounders.ConclusionsWe have shown that racial bias can exist in DL-based cine CMR segmentation models. We believe that this bias is due to the imbalanced nature of the training data (combined with physiological differences). This is supported by the results which show racial bias but not sex bias when trained using the UK Biobank database, which is sex-balanced but not race-balanced.Condensed AbstractAI algorithms have the potential to reflect or exacerbate racial/sex disparities in healthcare. We aimed to determine the impact of sex and race on the performance of an AI segmentation model for automatic CMR quantification in a cohort of 5,903 subjects from the UK Biobank database, which is sex-balanced but not race-balanced. We tested the model’s bias in performance using Dice scores and absolute/relative errors in measurements of biventricular volumes and function. Our study demonstrates that the model had a racial bias but no sex bias, and that subject characteristics and co-morbidities could not explain this bias.

One of the challenges in developing deep learning algorithms for medical image segmentation is the scarcity of annotated training data. To overcome this limitation, data augmentation and semi-supervised learning (SSL) methods have been developed. However, these methods have limited effectiveness as they either exploit the existing data set only (data augmentation) or risk negative impact by adding poor training examples (SSL). Segmentations are rarely the final product of medical image analysis - they are typically used in downstream tasks to infer higher-order patterns to evaluate diseases. Clinicians take into account a wealth of prior knowledge on biophysics and physiology when evaluating image analysis results. We have used these clinical assessments in previous works to create robust quality-control (QC) classifiers for automated cardiac magnetic resonance (CMR) analysis. In this paper, we propose a novel scheme that uses QC of the downstream task to identify high quality outputs of CMR segmentation networks, that are subsequently utilised for further network training. In essence, this provides quality-aware augmentation of training data in a variant of SSL for segmentation networks (semiQCSeg). We evaluate our approach in two CMR segmentation tasks (aortic and short axis cardiac volume segmentation) using UK Biobank data and two commonly used network architectures (U-net and a Fully Convolutional Network) and compare against supervised and SSL strategies. We show that semiQCSeg improves training of the segmentation networks. It decreases the need for labelled data, while outperforming the other methods in terms of Dice and clinical metrics. SemiQCSeg can be an efficient approach for training segmentation networks for medical image data when labelled datasets are scarce., MICCAI STACOM 2020

Background: Focal ventricular tachycardia (VT) is a life-threating arrhythmia, responsible for high morbidity rates and sudden cardiac death (SCD). Radiofrequency ablation is the only curative therapy against incessant VT; however, its success is dependent on accurate localization of its source, which is highly invasive and time-consuming.Objective: The goal of our study is, as a proof of concept, to demonstrate the possibility of utilizing electrogram (EGM) recordings from cardiac implantable electronic devices (CIEDs). To achieve this, we utilize fast and accurate whole torso electrophysiological (EP) simulations in conjunction with convolutional neural networks (CNNs) to automate the localization of focal VTs using simulated EGMs.Materials and Methods: A highly detailed 3D torso model was used to simulate ∼4000 focal VTs, evenly distributed across the left ventricle (LV), utilizing a rapid reaction-eikonal environment. Solutions were subsequently combined with lead field computations on the torso to derive accurate electrocardiograms (ECGs) and EGM traces, which were used as inputs to CNNs to localize focal sources. We compared the localization performance of a previously developed CNN architecture (Cartesian probability-based) with our novel CNN algorithm utilizing universal ventricular coordinates (UVCs).Results: Implanted device EGMs successfully localized VT sources with localization error (8.74 mm) comparable to ECG-based localization (6.69 mm). Our novel UVC CNN architecture outperformed the existing Cartesian probability-based algorithm (errors = 4.06 mm and 8.07 mm for ECGs and EGMs, respectively). Overall, localization was relatively insensitive to noise and changes in body compositions; however, displacements in ECG electrodes and CIED leads caused performance to decrease (errors 16–25 mm).Conclusion: EGM recordings from implanted devices may be used to successfully, and robustly, localize focal VT sources, and aid ablation planning.

Objective: Splenomegaly (abnormal splenic enlargement) is a potentially life-threatening condition that occurs in a range of clinical scenarios, including in patients suffering from Sickle cell disease (SCD). Therefore, spleen size assessments from ultrasound imaging are commonly performed in SCD clinics, and typically involve measuring the length of the spleen. However, the current workflow is prone to intra- and inter-observer variability and is dependent on the experience of the sonographer. Our objective was to automate the spleen length measurement process. Methods: Two deep learning-based approaches were investigated to achieve automated spleen length measurement from ultrasound images. One is a segmentation-based approach, where we trained a modified U-Net to obtain a spleen segmentation and then applied post-processing to measure the spleen length from the segmentation. The second approach is based on direct regression of spleen length. We also incorporated a quality control (QC) model to help less experienced sonographers ensure the quality of ultrasound images before measurement. Results: Our best model (segmentation-based approach) reached a mean percentage length error (MPLE) of 4.58% on good quality images, which is within the range of human expert inter-observer variability (5.78%). After including bad quality images, the incorporation of the QC step resulted in a significant reduction in MPLE (from 5.76% to 4.88%). Conclusion: Automated, quality-controlled spleen length measurement from ultrasound has been achieved with human-level accuracy. Significance: This proposed framework has the potential to assist in making robust and accurate assessments of the spleen, especially in settings where there is a lack of experienced sonographers.

Advances in deep learning (DL) have resulted in impressive accuracy in some medical image classification tasks, but often deep models lack interpretability. The ability of these models to explain their decisions is important for fostering clinical trust and facilitating clinical translation. Furthermore, for many problems in medicine there is a wealth of existing clinical knowledge to draw upon, which may be useful in generating explanations, but it is not obvious how this knowledge can be encoded into DL models - most models are learnt either from scratch or using transfer learning from a different domain. In this paper we address both of these issues. We propose a novel DL framework for image-based classification based on a variational autoencoder (VAE). The framework allows prediction of the output of interest from the latent space of the autoencoder, as well as visualisation (in the image domain) of the effects of crossing the decision boundary, thus enhancing the interpretability of the classifier. Our key contribution is that the VAE disentangles the latent space based on `explanations' drawn from existing clinical knowledge. The framework can predict outputs as well as explanations for these outputs, and also raises the possibility of discovering new biomarkers that are separate (or disentangled) from the existing knowledge. We demonstrate our framework on the problem of predicting response of patients with cardiomyopathy to cardiac resynchronization therapy (CRT) from cine cardiac magnetic resonance images. The sensitivity and specificity of the proposed model on the task of CRT response prediction are 88.43% and 84.39% respectively, and we showcase the potential of our model in enhancing understanding of the factors contributing to CRT response., Comment: MICCAI 2020 conference

With respect to spatial overlap, CNN-based segmentation of short axis cardiovascular magnetic resonance (CMR) images has achieved a level of performance consistent with inter observer variation. However, conventional training procedures frequently depend on pixel-wise loss functions, limiting optimisation with respect to extended or global features. As a result, inferred segmentations can lack spatial coherence, including spurious connected components or holes. Such results are implausible, violating the anticipated topology of image segments, which is frequently known a priori. Addressing this challenge, published work has employed persistent homology, constructing topological loss functions for the evaluation of image segments against an explicit prior. Building a richer description of segmentation topology by considering all possible labels and label pairs, we extend these losses to the task of multi-class segmentation. These topological priors allow us to resolve all topological errors in a subset of 150 examples from the ACDC short axis CMR training data set, without sacrificing overlap performance., To be presented at the STACOM workshop at MICCAI 2020

Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Advancing Impact Award scheme of the EPSRC Impact Acceleration Account at King’s College London Background Artificial intelligence (AI) has the potential to facilitate the automation of CMR analysis for biomarker extraction. However, most AI algorithms are trained on a specific input domain (e.g., scanner vendor or hospital-tailored imaging protocol) and lack the robustness to perform optimally when applied to CMR data from other input domains. Purpose To develop and validate a robust CMR analysis tool for automatic segmentation and cardiac function analysis which achieves state-of-the-art performance for multi-vendor short-axis cine CMR images. Methods The current work is an extension of our previously published quality-controlled AI-based tool for cine CMR analysis [1]. We deployed an AI algorithm that is equipped to handle different image sizes and domains automatically - the ‘nnU-Net’ framework [2] - and retrained our tool using the UK Biobank (UKBB) cohort population (n = 4,872) and a large database of clinical CMR studies obtained from two NHS hospitals (n = 3,406). The NHS hospital data came from three different scanner types: Siemens Aera 1.5T (n = 1,419), Philips Achieva 1.5T and 3T (n = 1,160), and Philips Ingenia 1.5T (n = 827). The ‘nnU-net’ was used to segment both ventricles and the myocardium. The proposed method was evaluated on randomly selected test sets from UKBB (n = 488) and NHS (n = 331) and on two external publicly available databases of clinical CMRs acquired on Philips, Siemens, General Electric (GE), and Canon CMR scanners – ACDC (n = 100) [3] and M&Ms (n = 321) [4]. We calculated the Dice scores - which measure the overlap between manual and automatic segmentations - and compared manual vs AI-based measures of biventricular volumes and function. Results Table 1 shows that the Dice scores for the NHS, ACDC, and M&Ms scans are similar to those obtained in the highly controlled, single vendor and single field strength UKBB scans. Although our AI-based tool was only trained on CMR scans from two vendors (Philips and Siemens), it performs similarly in unseen vendors (GE and Canon). Furthermore, it achieves state-of-the-art performance in online segmentation challenges, without being specifically trained on these databases. Table 1 also shows good agreement between manual and automated clinical measures of ejection fraction and ventricular volume and mass. Conclusions We show that our proposed AI-based tool, which combines training on a large-scale multi-domain CMR database with a state-of-the-art AI algorithm, allows us to robustly deal with routine clinical data from multiple centres, vendors, and field strengths. This is a fundamental step for the clinical translation of AI algorithms. Moreover, our method yields a range of additional metrics of cardiac function (filling and ejection rates, regional wall motion, and strain) at no extra computational cost.

Highlights • A method to segment the vocal tract and articulators in MR images was developed. • Median accuracy: Dice coefficient of 0.92; general Hausdorff distance of 5mm. • Developed to facilitate quantitative analysis of the vocal tract and articulators. • Intended for use in clinical and non-clinical studies of speech. • A novel clinically relevant segmentation accuracy metric was also developed., Background and Objective Magnetic resonance (MR) imaging is increasingly used in studies of speech as it enables non-invasive visualisation of the vocal tract and articulators, thus providing information about their shape, size, motion and position. Extraction of this information for quantitative analysis is achieved using segmentation. Methods have been developed to segment the vocal tract, however, none of these also fully segment any articulators. The objective of this work was to develop a method to fully segment multiple groups of articulators as well as the vocal tract in two-dimensional MR images of speech, thus overcoming the limitations of existing methods. Methods Five speech MR image sets (392 MR images in total), each of a different healthy adult volunteer, were used in this work. A fully convolutional network with an architecture similar to the original U-Net was developed to segment the following six regions in the image sets: the head, soft palate, jaw, tongue, vocal tract and tooth space. A five-fold cross-validation was performed to investigate the segmentation accuracy and generalisability of the network. The segmentation accuracy was assessed using standard overlap-based metrics (Dice coefficient and general Hausdorff distance) and a novel clinically relevant metric based on velopharyngeal closure. Results The segmentations created by the method had a median Dice coefficient of 0.92 and a median general Hausdorff distance of 5mm. The method segmented the head most accurately (median Dice coefficient of 0.99), and the soft palate and tooth space least accurately (median Dice coefficients of 0.92 and 0.93 respectively). The segmentations created by the method correctly showed 90% (27 out of 30) of the velopharyngeal closures in the MR image sets. Conclusions An automatic method to fully segment multiple groups of articulators as well as the vocal tract in two-dimensional MR images of speech was successfully developed. The method is intended for use in clinical and non-clinical speech studies which involve quantitative analysis of the shape, size, motion and position of the vocal tract and articulators. In addition, a novel clinically relevant metric for assessing the accuracy of vocal tract and articulator segmentation methods was developed.

We present a novel multimodal deep learning framework for cardiac resynchronisation therapy (CRT) response prediction from 2D echocardiography and cardiac magnetic resonance (CMR) data. The proposed method first uses the `nnU-Net' segmentation model to extract segmentations of the heart over the full cardiac cycle from the two modalities. Next, a multimodal deep learning classifier is used for CRT response prediction, which combines the latent spaces of the segmentation models of the two modalities. At inference time, this framework can be used with 2D echocardiography data only, whilst taking advantage of the implicit relationship between CMR and echocardiography features learnt from the model. We evaluate our pipeline on a cohort of 50 CRT patients for whom paired echocardiography/CMR data were available, and results show that the proposed multimodal classifier results in a statistically significant improvement in accuracy compared to the baseline approach that uses only 2D echocardiography data. The combination of multimodal data enables CRT response to be predicted with 77.38% accuracy (83.33% sensitivity and 71.43% specificity), which is comparable with the current state-of-the-art in machine learning-based CRT response prediction. Our work represents the first multimodal deep learning approach for CRT response prediction.

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia, characterised by a rapid and irregular electrical activation of the atria. Treatments for AF are often ineffective and few atrial biomarkers exist to automatically characterise atrial function and aid in treatment selection for AF. Clinical metrics of left atrial (LA) function, such as ejection fraction (EF) and active atrial contraction ejection fraction (aEF), are promising, but have until now typically relied on volume estimations extrapolated from single-slice images.

The subject of ‘fairness’ in artificial intelligence (AI) refers to assessing AI algorithms for potential bias based on demographic characteristics such as race and gender, and the development of algorithms to address this bias. Most applications to date have been in computer vision, although some work in healthcare has started to emerge. The use of deep learning (DL) in cardiac MR segmentation has led to impressive results in recent years, and such techniques are starting to be translated into clinical practice. However, no work has yet investigated the fairness of such models. In this work, we perform such an analysis for racial/gender groups, focusing on the problem of training data imbalance, using a nnU-Net model trained and evaluated on cine short axis cardiac MR data from the UK Biobank dataset, consisting of 5,903 subjects from 6 different racial groups. We find statistically significant differences in Dice performance between different racial groups. To reduce the racial bias, we investigated three strategies: (1) stratified batch sampling, in which batch sampling is stratified to ensure balance between racial groups; (2) fair meta-learning for segmentation, in which a DL classifier is trained to classify race and jointly optimized with the segmentation model; and (3) protected group models, in which a different segmentation model is trained for each racial group. We also compared the results to the scenario where we have a perfectly balanced database. To assess fairness we used the standard deviation (SD) and skewed error ratio (SER) of the average Dice values. Our results demonstrate that the racial bias results from the use of imbalanced training data, and that all proposed bias mitigation strategies improved fairness, with the best SD and SER resulting from the use of protected group models.

Multi-class segmentation of cardiac magnetic resonance (CMR) images seeks a separation of data into anatomical components with known structure and configuration. The most popular CNN-based methods are optimised using pixel wise loss functions, ignorant of the spatially extended features that characterise anatomy. Therefore, whilst sharing a high spatial overlap with the ground truth, inferred CNN-based segmentations can lack coherence, including spurious connected components, holes and voids. Such results are implausible, violating anticipated anatomical topology. In response, (single-class) persistent homology-based loss functions have been proposed to capture global anatomical features. Our work extends these approaches to the task of multi-class segmentation. Building an enriched topological description of all class labels and class label pairs, our loss functions make predictable and statistically significant improvements in segmentation topology using a CNN-based post-processing framework. We also present (and make available) a highly efficient implementation based on cubical complexes and parallel execution, enabling practical application within high resolution 3D data for the first time. We demonstrate our approach on 2D short axis and 3D whole heart CMR segmentation, advancing a detailed and faithful analysis of performance on two publicly available datasets., Comment: Version accepted for publication in IEEE Transactions on Medical Imaging

To date, regional atrial strains have not been imaged in vivo, despite their potential to provide useful clinical information. To address this gap, we present a novel CINE MRI protocol capable of imaging the entire left atrium at an isotropic 2-mm resolution in one single breath-hold. As proof of principle, we acquired data in 10 healthy volunteers and 2 cardiovascular patients using this technique. We also demonstrated how regional atrial strains can be estimated from this data following a manual segmentation of the left atrium using automatic image tracking techniques. The estimated principal strains vary smoothly across the left atrium and have a similar magnitude to estimates reported in the literature.

Sickle Cell Disease (SCD) is one of the most common genetic diseases in the world. Splenomegaly (abnormal enlargement of the spleen) is frequent among children with SCD. If left untreated, splenomegaly can be life-threatening. The current workflow to measure spleen size includes palpation, possibly followed by manual length measurement in 2D ultrasound imaging. However, this manual measurement is dependent on operator expertise and is subject to intra- and inter-observer variability. We investigate the use of deep learning to perform automatic estimation of spleen length from ultrasound images. We investigate two types of approach, one segmentation-based and one based on direct length estimation, and compare the results against measurements made by human experts. Our best model (segmentation-based) achieved a percentage length error of 7.42%, which is approaching the level of inter-observer variability (5.47%–6.34%). To the best of our knowledge, this is the first attempt to measure spleen size in a fully automated way from ultrasound images.

Maintaining good cardiac function for as long as possible is a major concern for healthcare systems worldwide and there is much interest in learning more about the impact of different risk factors on cardiac health. The aim of this study is to analyze the impact of systolic blood pressure (SBP) on cardiac function while preserving the interpretability of the model using known clinical biomarkers in a large cohort of the UK Biobank population. We propose a novel framework that combines deep learning based estimation of interpretable clinical biomarkers from cardiac cine MR data with a variational autoencoder (VAE). The VAE architecture integrates a regression loss in the latent space, which enables the progression of cardiac health with SBP to be learnt. Results on 3,600 subjects from the UK Biobank show that the proposed model allows us to gain important insight into the deterioration of cardiac function with increasing SBP, identify key interpretable factors involved in this process, and lastly exploit the model to understand patterns of positive and adverse adaptation of cardiac function.

Segmenting anatomical structures in medical images has been successfully addressed with deep learning methods for a range of applications. However, this success is heavily dependent on the quality of the image that is being segmented. A commonly neglected point in the medical image analysis community is the vast amount of clinical images that have severe image artefacts due to organ motion, movement of the patient and/or image acquisition related issues. In this paper, we discuss the implications of image motion artefacts on cardiac MR segmentation and compare a variety of approaches for jointly correcting for artefacts and segmenting the cardiac cavity. The method is based on our recently developed joint artefact detection and reconstruction method, which reconstructs high quality MR images from k-space using a joint loss function and essentially converts the artefact correction task to an under-sampled image reconstruction task by enforcing a data consistency term. In this paper, we propose to use a segmentation network coupled with this in an end-to-end framework. Our training optimises three different tasks: 1) image artefact detection, 2) artefact correction and 3) image segmentation. We train the reconstruction network to automatically correct for motion-related artefacts using synthetically corrupted cardiac MR k-space data and uncorrected reconstructed images. Using a test set of 500 2D+time cine MR acquisitions from the UK Biobank data set, we achieve demonstrably good image quality and high segmentation accuracy in the presence of synthetic motion artefacts. We showcase better performance compared to various image correction architectures.

The neuropathological confirmation of amyloid-β (Aβ) plaques and tau neurofibrillary tangles (NFT) remains the gold standard for a definitive diagnosis of Alzheimer's disease (AD). Nowadays, the in vivo diagnosis of AD is greatly aided by both cerebrospinal fluid (CSF) and positron emission tomography (PET) biomarkers. Although highly accurate, their broad implementation is restricted by high cost, limited accessibility and invasiveness. We recently developed a high-performance, ultrasensitive immunoassay for the quantification of tau phosphorylated at threonine-181 (p-tau181) in plasma, which identifies AD pathophysiology with high accuracy. However, it remains unclear whether plasma p-tau181, measured years before the death, can predict the eventual neuropathological confirmation of AD, and successfully discriminates AD from non-AD dementia pathologies. We studied a unique cohort of 115 individuals with longitudinal blood collections with clinical evaluation at 8, 4 and 2 years prior to neuropathological assessment at death. The results demonstrate that plasma p-tau181 associates better with AD neuropathology and Braak staging than a clinical diagnosis 8 years before post-mortem. Moreover, while all patients had a diagnosis of AD dementia during life, plasma p-tau181 proved to discriminate AD from non-AD pathologies with high accuracy (AUC = 97.4%, 95% CI = 94.1-100%) even 8 years before death. Additionally, the longitudinal trajectory of plasma p-tau181 was assessed in all patients. We found that the main increases in plasma p-tau181 occurred between 8 and 4 years prior to death in patients with AD neuropathology and later plateauing. In contrast, non-AD pathologies and controls exhibited minor, albeit significant, increases in p-tau181 up until death. Overall, our study demonstrates that plasma p-tau181 is highly predictive and specific of AD neuropathology years before post-mortem examination. These data add further support for the use of plasma p-tau181 to aid clinical management in primary care and recruitment for clinical trials. Open access funding provided by University of Gothenburg. This study represents independent research partly funded by the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. Tissue samples were supplied by The London Neurodegenerative Diseases Brain Bank, which receives funding from the UK Medical Research Council and as part of the Brains for Dementia Research programme, jointly funded by Alzheimer’s Research UK and the Alzheimer’s Society. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health. The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. TKK holds a postdoctoral fellowship from the BrightFocus Foundation (#A2020812F), and was further supported by the Swedish Alzheimer Foundation (Alzheimerfonden; #AF-930627), the Swedish Brain Foundation (Hjärnfonden; #FO2020-0240), the Swedish Dementia Foundation (Demensförbundet), the Agneta Prytz-Folkes and Gösta Folkes Foundation, Gamla Tjänarinnor, the Aina (Ann) Wallströms and Mary-Ann Sjöbloms Foundation, the Gun and Bertil Stohnes Foundation, and the Anna Lisa and Brother Björnsson’s Foundation. MSC received funding from the European Union’s Horizon 2020 Research and Innovation Program under the Marie Sklodowska-Curie action grant agreement no. 752310, and currently receives funding from Instituto de Salud Carlos III (PI19/00155) and from the Spanish Ministry of Science, Innovation and Universities (Juan de la Cierva Programme grant IJC2018-037478-I). AH is funded by Research Centre for Mental Health and Biomedical Research Unit for Dementia. KB holds the Torsten Söderberg Professorship in Medicine and is supported by grants from the Swedish Research Council, the Swedish Alzheimer Foundation, and the Swedish Brain Foundation. AH is funded by Research Centre for Mental Health and Biomedical Research Unit for Dementia. KB holds the Torsten Söderberg Professorship in Medicine and is supported by grants from the Swedish Research Council, the Swedish Alzheimer Foundation, and the Swedish Brain Foundation. KB holds the Torsten Söderberg Professorship in Medicine at the Royal Swedish Academy of Sciences, and is supported by the Swedish Research Council (#2017-00915), the Swedish Alzheimer Foundation (#AF-742881), Hjärnfonden, Sweden (#FO2017-0243), and a grant (#ALFGBG-715986) from the Swedish state under the agreement between the Swedish government and the County Councils, the ALF-agreement. NJA is supported by the Wallenberg Centre for Molecular and Translational Medicine, the Swedish Alzheimer Foundation (Alzheimerfonden), the Swedish Dementia Foundation (Demensförbundet), and Hjärnfonden, Sweden.

The goal of this work is to train a neural network which approximates solutions to the Navier-Stokes equations across a region of parameter space, in which the parameters define physical properties such as domain shape and boundary conditions. The contributions of this work are threefold: 1) To demonstrate that neural networks can be efficient aggregators of whole families of parameteric solutions to physical problems, trained using data created with traditional, trusted numerical methods such as finite elements. Advantages include extremely fast evaluation of pressure and velocity at any point in physical and parameter space (asymptotically, ~3 $\mu s$ / query), and data compression (the network requires 99\% less storage space compared to its own training data). 2) To demonstrate that the neural networks can accurately interpolate between finite element solutions in parameter space, allowing them to be instantly queried for pressure and velocity field solutions to problems for which traditional simulations have never been performed. 3) To introduce an active learning algorithm, so that during training, a finite element solver can automatically be queried to obtain additional training data in locations where the neural network's predictions are in most need of improvement, thus autonomously acquiring and efficiently distributing training data throughout parameter space. In addition to the obvious utility of Item 2, above, we demonstrate an application of the network in rapid parameter sweeping, very precisely predicting the degree of narrowing in a tube which would result in a 50\% increase in end-to-end pressure difference at a given flow rate. This capability could have applications in both medical diagnosis of arterial disease, and in computer-aided design., Comment: 16 pages, 9 figures; added missing details from author affiliations

We present a framework for combining a cardiac motion atlas with non-motion data. The atlas represents cardiac cycle motion across a number of subjects in a common space based on rich motion descriptors capturing 3D displacement, velocity, strain and strain rate. The non-motion data are derived from a variety of sources such as imaging, electrocardiogram (ECG) and clinical reports. Once in the atlas space, we apply a novel supervised learning approach based on random projections and ensemble learning to learn the relationship between the atlas data and some desired clinical output. We apply our framework to the problem of predicting response to Cardiac Resynchronisation Therapy (CRT). Using a cohort of 34 patients selected for CRT using conventional criteria, results show that the combination of motion and non-motion data enables CRT response to be predicted with 91.2% accuracy (100% sensitivity and 62.5% specificity), which compares favourably with the current state-of-the-art in CRT response prediction.

Background Ex-vivo cardiovascular magnetic resonance (CMR) imaging has played an important role in the validation of in-vivo CMR characterization of pathological processes. However, comparison between in-vivo and ex-vivo imaging remains challenging due to shape changes occurring between the two states, which may be non-uniform across the diseased heart. A novel two-step process to facilitate registration between ex-vivo and in-vivo CMR was developed and evaluated in a porcine model of chronic myocardial infarction (MI). Methods Seven weeks after ischemia-reperfusion MI, 12 swine underwent in-vivo CMR imaging with late gadolinium enhancement followed by ex-vivo CMR 1 week later. Five animals comprised the control group, in which ex-vivo imaging was undertaken without any support in the LV cavity, 7 animals comprised the experimental group, in which a two-step registration optimization process was undertaken. The first step involved a heart specific flexible 3D printed scaffold generated from in-vivo CMR, which was used to maintain left ventricular (LV) shape during ex-vivo imaging. In the second step, a non-rigid co-registration algorithm was applied to align in-vivo and ex-vivo data. Tissue dimension changes between in-vivo and ex-vivo imaging were compared between the experimental and control group. In the experimental group, tissue compartment volumes and thickness were compared between in-vivo and ex-vivo data before and after non-rigid registration. The effectiveness of the alignment was assessed quantitatively using the DICE similarity coefficient. Results LV cavity volume changed more in the control group (ratio of cavity volume between ex-vivo and in-vivo imaging in control and experimental group 0.14 vs 0.56, p p p = 0.034). Following the non-rigid co-registration step of the process, the DICE similarity coefficients for the myocardium, LV cavity and scar were 0.93 (±0.02), 0.89 (±0.01) and 0.77 (±0.07) respectively and the myocardial tissue and LV cavity volumes had a ratio of 1.03 and 1.00 respectively. Conclusions The pattern of the morphological changes seen between the in-vivo and the ex-vivo LV differs between scar and healthy myocardium. A 3D printed flexible scaffold based on the in-vivo shape of the LV cavity is an effective strategy to minimize morphological changes in the ex-vivo LV. The subsequent non-rigid registration step further improved the co-registration and local comparison between in-vivo and ex-vivo data.

Motion imaging phantoms are expensive, bulky and difficult to transport and set-up. The purpose of this paper is to demonstrate a simple approach to the design of multi-modality motion imaging phantoms that use mechanically stored energy to produce motion. We propose two phantom designs that use mainsprings and elastic bands to store energy. A rectangular piece was attached to an axle at the end of the transmission chain of each phantom, and underwent a rotary motion upon release of the mechanical motor. The phantoms were imaged with MRI and US, and the image sequences were embedded in a 1D non linear manifold (Laplacian Eigenmap) and the spectrogram of the embedding was used to derive the angular velocity over time. The derived velocities were consistent and reproducible within a small error. The proposed motion phantom concept showed great potential for the construction of simple and affordable motion phantoms, Comment: Accepted for publication at IEEE EMBC (41st International Engineering in Medicine and Biology Conference) 2019

This study sought to develop a fully automated framework for cardiac function analysis from cardiac magnetic resonance (CMR), including comprehensive quality control (QC) algorithms to detect erroneous output.Analysis of cine CMR imaging using deep learning (DL) algorithms could automate ventricular function assessment. However, variable image quality, variability in phenotypes of disease, and unavoidable weaknesses in training of DL algorithms currently prevent their use in clinical practice.The framework consists of a pre-analysis DL image QC, followed by a DL algorithm for biventricular segmentation in long-axis and short-axis views, myocardial feature-tracking (FT), and a post-analysis QC to detect erroneous results. The study validated the framework in healthy subjects and cardiac patients by comparison against manual analysis (n = 100) and evaluation of the QC steps' ability to detect erroneous results (n = 700). Next, this method was used to obtain reference values for cardiac function metrics from the UK Biobank.Automated analysis correlated highly with manual analysis for left and right ventricular volumes (all r0.95), strain (circumferential r = 0.89, longitudinal r0.89), and filling and ejection rates (all r ≥ 0.93). There was no significant bias for cardiac volumes and filling and ejection rates, except for right ventricular end-systolic volume (bias +1.80 ml; p = 0.01). The bias for FT strain was 1.3%. The sensitivity of detection of erroneous output was 95% for volume-derived parameters and 93% for FT strain. Finally, reference values were automatically derived from 2,029 CMR exams in healthy subjects.The study demonstrates a DL-based framework for automated, quality-controlled characterization of cardiac function from cine CMR, without the need for direct clinician oversight.

Deep learning methods for classifying medical images have demonstrated impressive accuracy in a wide range of tasks but often these models are hard to interpret, limiting their applicability in clinical practice. In this work we introduce a convolutional neural network model for identifying disease in temporal sequences of cardiac MR segmentations which is interpretable in terms of clinically familiar measurements. The model is based around a variational autoencoder, reducing the input into a low-dimensional latent space in which classification occurs. We then use the recently developed ‘concept activation vector’ technique to associate concepts which are diagnostically meaningful (eg. clinical biomarkers such as ‘low left-ventricular ejection fraction’) to certain vectors in the latent space. These concepts are then qualitatively inspected by observing the change in the image domain resulting from interpolations in the latent space in the direction of these vectors. As a result, when the model classifies images it is also capable of providing naturally interpretable concepts relevant to that classification and demonstrating the meaning of those concepts in the image domain. Our approach is demonstrated on the UK Biobank cardiac MRI dataset where we detect the presence of coronary artery disease.

Patient-specific 3D printing of congenital heart anatomy demands an accurate segmentation of the thin tissue interfaces which characterise these diagnoses. Even when a label set has a high spatial overlap with the ground truth, inaccurate delineation of these interfaces can result in topological errors. These compromise the clinical utility of such models due to the anomalous appearance of defects. CNNs have achieved state-of-the-art performance in segmentation tasks. Whilst data augmentation has often played an important role, we show that conventional image resampling schemes used therein can introduce topological changes in the ground truth labelling of augmented samples. We present a novel pipeline to correct for these changes, using a fast-marching algorithm to enforce the topology of the ground truth labels within their augmented representations. In so doing, we invoke the idea of cardiac contiguous topology to describe an arbitrary combination of congenital heart defects and develop an associated, clinically meaningful metric to measure the topological correctness of segmentations. In a series of five-fold cross-validations, we demonstrate the performance gain produced by this pipeline and the relevance of topological considerations to the segmentation of congenital heart defects. We speculate as to the applicability of this approach to any segmentation task involving morphologically complex targets.