Non-invasive localization of post-infarct ventricular tachycardia exit sites to guide ablation planning: a computational deep learning platform utilizing the 12-lead electrocardiogram and intracardiac electrograms from implanted devices.

AIMS: Existing 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 RESULTS: A 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. CONCLUSION: The 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.

Clinical benefit of AI-assisted lung ultrasound in a resource-limited intensive care unit.

BACKGROUND: Interpreting point-of-care lung ultrasound (LUS) images from intensive care unit (ICU) patients can be challenging, especially in low- and middle- income countries (LMICs) where there is limited training available. Despite recent advances in the use of Artificial Intelligence (AI) to automate many ultrasound imaging analysis tasks, no AI-enabled LUS solutions have been proven to be clinically useful in ICUs, and specifically in LMICs. Therefore, we developed an AI solution that assists LUS practitioners and assessed its usefulness in  a low resource ICU. METHODS: This was a three-phase prospective study. In the first phase, the performance of four different clinical user groups in interpreting LUS clips was assessed. In the second phase, the performance of 57 non-expert clinicians with and without the aid of a bespoke AI tool for LUS interpretation was assessed in retrospective offline clips. In the third phase, we conducted a prospective study in the ICU where 14 clinicians were asked to carry out LUS examinations in 7 patients with and without our AI tool and we interviewed the clinicians regarding the usability of the AI tool. RESULTS: The average accuracy of beginners' LUS interpretation was 68.7% [95% CI 66.8-70.7%] compared to 72.2% [95% CI 70.0-75.6%] in intermediate, and 73.4% [95% CI 62.2-87.8%] in advanced users. Experts had an average accuracy of 95.0% [95% CI 88.2-100.0%], which was significantly better than beginners, intermediate and advanced users (p < 0.001). When supported by our AI tool for interpreting retrospectively acquired clips, the non-expert clinicians improved their performance from an average of 68.9% [95% CI 65.6-73.9%] to 82.9% [95% CI 79.1-86.7%], (p < 0.001). In prospective real-time testing, non-expert clinicians improved their baseline performance from 68.1% [95% CI 57.9-78.2%] to 93.4% [95% CI 89.0-97.8%], (p < 0.001) when using our AI tool. The time-to-interpret clips improved from a median of 12.1 s (IQR 8.5-20.6) to 5.0 s (IQR 3.5-8.8), (p < 0.001) and clinicians' median confidence level improved from 3 out of 4 to 4 out of 4 when using our AI tool. CONCLUSIONS: AI-assisted LUS can help non-expert clinicians in an LMIC ICU improve their performance in interpreting LUS features more accurately, more quickly and more confidently.

Automated quantification of myocardial tissue characteristics from native T1 mapping using neural networks with uncertainty-based quality-control.

BACKGROUND: Tissue characterisation with cardiovascular magnetic resonance (CMR) parametric mapping has the potential to detect and quantify both focal and diffuse alterations in myocardial structure not assessable by late gadolinium enhancement. Native T1 mapping in particular has shown promise as a useful biomarker to support diagnostic, therapeutic and prognostic decision-making in ischaemic and non-ischaemic cardiomyopathies. METHODS: Convolutional neural networks (CNNs) with Bayesian inference are a category of artificial neural networks which model the uncertainty of the network output. This study presents an automated framework for tissue characterisation from native shortened modified Look-Locker inversion recovery ShMOLLI T1 mapping at 1.5 T using a Probabilistic Hierarchical Segmentation (PHiSeg) network (PHCUMIS 119-127, 2019). In addition, we use the uncertainty information provided by the PHiSeg network in a novel automated quality control (QC) step to identify uncertain T1 values. The PHiSeg network and QC were validated against manual analysis on a cohort of the UK Biobank containing healthy subjects and chronic cardiomyopathy patients (N=100 for the PHiSeg network and N=700 for the QC). We used the proposed method to obtain reference T1 ranges for the left ventricular (LV) myocardium in healthy subjects as well as common clinical cardiac conditions. RESULTS: T1 values computed from automatic and manual segmentations were highly correlated (r=0.97). Bland-Altman analysis showed good agreement between the automated and manual measurements. The average Dice metric was 0.84 for the LV myocardium. The sensitivity of detection of erroneous outputs was 91%. Finally, T1 values were automatically derived from 11,882 CMR exams from the UK Biobank. For the healthy cohort, the mean (SD) corrected T1 values were 926.61 (45.26), 934.39 (43.25) and 927.56 (50.36) for global, interventricular septum and free-wall respectively. CONCLUSIONS: The proposed pipeline allows for automatic analysis of myocardial native T1 mapping and includes a QC process to detect potentially erroneous results. T1 reference values were presented for healthy subjects and common clinical cardiac conditions from the largest cohort to date using T1-mapping images.

Respiratory motion correction of PET using MR-constrained PET-PET registration.

BACKGROUND: Respiratory motion in positron emission tomography (PET) is an unavoidable source of error in the measurement of tracer uptake, lesion position and lesion size. The introduction of PET-MR dual modality scanners opens a new avenue for addressing this issue. Motion models offer a way to estimate motion using a reduced number of parameters. This can be beneficial for estimating motion from PET, which can otherwise be difficult due to the high level of noise of the data. METHOD: We propose a novel technique that makes use of a respiratory motion model, formed from initial MR scan data. The motion model is used to constrain PET-PET registrations between a reference PET gate and the gates to be corrected. For evaluation, PET with added FDG-avid lesions was simulated from real, segmented, ultrashort echo time MR data obtained from four volunteers. Respiratory motion was included in the simulations using motion fields derived from real dynamic 3D MR volumes obtained from the same volunteers. RESULTS: Performance was compared to an MR-derived motion model driven method (which requires constant use of the MR scanner) and to unconstrained PET-PET registration of the PET gates. Without motion correction, a median drop in uncorrected lesion [Formula: see text] intensity to [Formula: see text] and an increase in median head-foot lesion width, specified by a minimum bounding box, to [Formula: see text] was observed relative to the corresponding measures in motion-free simulations. The proposed method corrected these values to [Formula: see text] ([Formula: see text]) and [Formula: see text] ([Formula: see text]) respectively, with notably improved performance close to the diaphragm and in the liver. Median lesion displacement across all lesions was observed to be [Formula: see text] without motion correction, which was reduced to [Formula: see text] ([Formula: see text]) with motion correction. DISCUSSION: This paper presents a novel technique for respiratory motion correction of PET data in PET-MR imaging. After an initial 30 second MR scan, the proposed technique does not require use of the MR scanner for motion correction purposes, making it suitable for MR-intensive studies or sequential PET-MR. The accuracy of the proposed technique was similar to both comparative methods, but robustness was improved compared to the PET-PET technique, particularly in regions with higher noise such as the liver.

Comparison of deep-learning multimodality data fusion strategies in mandibular osteoradionecrosis NTCP modelling using clinical variables and radiation dose distribution volumes.

Objective.Normal tissue complication probability (NTCP) modelling is rapidly embracing deep learning (DL) methods, acknowledging the importance of spatial dose information. Finding effective ways to combine information from radiation dose distribution maps (dosiomics) and clinical data involves technical challenges and requires domain knowledge. We propose different multi-modality data fusion strategies to facilitate future DL-based NTCP studies.Approach.Early, joint and late DL multi-modality fusion strategies were compared using clinical and mandibular radiation dose distribution volumes. These were contrasted with single-modality models: a random forest trained on non-image data (clinical, demographic and dose-volume metrics) and a 3D DenseNet-40 trained on image data (mandibular dose distribution maps). The study involved a matched cohort of 92 osteoradionecrosis cases and 92 controls from a single institution.Main results.The late fusion model exhibited superior discrimination and calibration performance, while the join fusion achieved a more balanced distribution of the predicted probabilities. Discrimination performance did not significantly differ between strategies. Late fusion, though less technically complex, lacks crucial inter-modality interactions for NTCP modelling. In contrast, joint fusion, despite its complexity, resulted in a single network training process which included intra- and inter-modality interactions in its model parameter optimisation.Significance.This study is a pioneering effort in comparing different strategies for including image data into DL-based NTCP models in combination with lower dimensional data such as clinical variables. The discrimination performance of such multi-modality NTCP models and the choice of fusion strategy will depend on the distribution and quality of both types of data. Multiple data fusion strategies should be compared and reported in multi-modality NTCP modelling using DL.

Effect of synthetic CT on dose-derived toxicity predictors for MR-only prostate radiotherapy.

Objectives: Toxicity-driven adaptive radiotherapy (RT) is enhanced by the superior soft tissue contrast of magnetic resonance (MR) imaging compared with conventional computed tomography (CT). However, in an MR-only RT pathway synthetic CTs (sCT) are required for dose calculation. This study evaluates 3 sCT approaches for accurate rectal toxicity prediction in prostate RT. Methods: Thirty-six patients had MR (T2-weighted acquisition optimized for anatomical delineation, and T1-Dixon) with same day standard-of-care planning CT for prostate RT. Multiple sCT were created per patient using bulk density (BD), tissue stratification (TS, from T1-Dixon) and deep-learning (DL) artificial intelligence (AI) (from T2-weighted) approaches for dose distribution calculation and creation of rectal dose volume histograms (DVH) and dose surface maps (DSM) to assess grade-2 (G2) rectal bleeding risk. Results: Maximum absolute errors using sCT for DVH-based G2 rectal bleeding risk (risk range 1.6% to 6.1%) were 0.6% (BD), 0.3% (TS) and 0.1% (DL). DSM-derived risk prediction errors followed a similar pattern. DL sCT has voxel-wise density generated from T2-weighted MR and improved accuracy for both risk-prediction methods. Conclusions: DL improves dosimetric and predicted risk calculation accuracy. Both TS and DL methods are clinically suitable for sCT generation in toxicity-guided RT, however, DL offers increased accuracy and offers efficiencies by removing the need for T1-Dixon MR. Advances in knowledge: This study demonstrates novel insights regarding the effect of sCT on predictive toxicity metrics, demonstrating clear accuracy improvement with increased sCT resolution. Accuracy of toxicity calculation in MR-only RT should be assessed for all treatment sites where dose to critical structures will guide adaptive-RT strategies. Clinical trial registration number: Patient data were taken from an ethically approved (UK Health Research Authority) clinical trial run at Guy's and St Thomas' NHS Foundation Trust. Study Name: MR-simulation in Radiotherapy for Prostate Cancer. ClinicalTrials.gov Identifier: NCT03238170.

Clinical evaluation of AI-assisted muscle ultrasound for monitoring muscle wasting in ICU patients.

Muscle ultrasound has been shown to be a valid and safe imaging modality to assess muscle wasting in critically ill patients in the intensive care unit (ICU). This typically involves manual delineation to measure the rectus femoris cross-sectional area (RFCSA), which is a subjective, time-consuming, and laborious task that requires significant expertise. We aimed to develop and evaluate an AI tool that performs automated recognition and measurement of RFCSA to support non-expert operators in measurement of the RFCSA using muscle ultrasound. Twenty patients were recruited between Feb 2023 and July 2023 and were randomized sequentially to operators using AI (n = 10) or non-AI (n = 10). Muscle loss during ICU stay was similar for both methods: 26 ± 15% for AI and 23 ± 11% for the non-AI, respectively (p = 0.13). In total 59 ultrasound examinations were carried out (30 without AI and 29 with AI). When assisted by our AI tool, the operators showed less variability between measurements with higher intraclass correlation coefficients (ICCs 0.999 95% CI 0.998-0.999 vs. 0.982 95% CI 0.962-0.993) and lower Bland Altman limits of agreement (± 1.9% vs. ± 6.6%) compared to not using the AI tool. The time spent on scans reduced significantly from a median of 19.6 min (IQR 16.9-21.7) to 9.4 min (IQR 7.2-11.7) compared to when using the AI tool (p < 0.001). AI-assisted muscle ultrasound removes the need for manual tracing, increases reproducibility and saves time. This system may aid monitoring muscle size in ICU patients assisting rehabilitation programmes.

A segmentation-informed deep learning framework to register dynamic two-dimensional magnetic resonance images of the vocal tract during speech.

Objective: Dynamic 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. Methods: A 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. Results: The 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. Conclusions: A framework was successfully developed and found to more accurately estimate articulator motion than five current state-of-the-art methods and frameworks. Significance: The first deep-learning-based framework specifically for registering dynamic two-dimensional MR images of the vocal tract during speech has been developed and evaluated.

An Investigation into Race Bias in Random Forest Models Based on Breast DCE-MRI Derived Radiomics Features.

Recent research has shown that artificial intelligence (AI) models can exhibit bias in performance when trained using data that are imbalanced by protected attribute(s). Most work to date has focused on deep learning models, but classical AI techniques that make use of hand-crafted features may also be susceptible to such bias. In this paper we investigate the potential for race bias in random forest (RF) models trained using radiomics features. Our application is prediction of tumour molecular subtype from dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) of breast cancer patients. Our results show that radiomics features derived from DCE-MRI data do contain race-identifiable information, and that RF models can be trained to predict White and Black race from these data with 60-70% accuracy, depending on the subset of features used. Furthermore, RF models trained to predict tumour molecular subtype using race-imbalanced data seem to produce biased behaviour, exhibiting better performance on test data from the race on which they were trained.

A Persistent Homology-Based Topological Loss for CNN-Based Multiclass Segmentation of CMR.

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.

Addressing Deep Learning Model Calibration Using Evidential Neural Networks And Uncertainty-Aware Training.

In terms of accuracy, deep learning (DL) models have had considerable success in classification problems for medical imaging applications. However, it is well-known that the outputs of such models, which typically utilise the SoftMax function in the final classification layer can be over-confident, i.e. they are poorly calibrated. Two competing solutions to this problem have been proposed: uncertainty-aware training and evidential neural networks (ENNs). In this paper we perform an investigation into the improvements to model calibration that can be achieved by each of these approaches individually, and their combination. We perform experiments on two classification tasks: a simpler MNIST digit classification task and a more complex and realistic medical imaging artefact detection task using Phase Contrast Cardiac Magnetic Resonance images. The experimental results demonstrate that model calibration can suffer when the task becomes challenging enough to require a higher capacity model. However, in our complex artefact detection task we saw an improvement in calibration for both a low and higher capacity model when implementing both the ENN and uncertainty-aware training together, indicating that this approach can offer a promising way to improve calibration in such settings. The findings highlight the potential use of these approaches to improve model calibration in a complex application, which would in turn improve clinician trust in DL models.

Exploring interpretability in deep learning prediction of successful ablation therapy for atrial fibrillation.

Background: Radiofrequency catheter ablation (RFCA) therapy is the first-line treatment for atrial fibrillation (AF), the most common type of cardiac arrhythmia globally. However, the procedure currently has low success rates in dealing with persistent AF, with a reoccurrence rate of ∼50% post-ablation. Therefore, deep learning (DL) has increasingly been applied to improve RFCA treatment for AF. However, for a clinician to trust the prediction of a DL model, its decision process needs to be interpretable and have biomedical relevance. Aim: This study explores interpretability in DL prediction of successful RFCA therapy for AF and evaluates if pro-arrhythmogenic regions in the left atrium (LA) were used in its decision process. Methods: AF and its termination by RFCA have been simulated in MRI-derived 2D LA tissue models with segmented fibrotic regions (n = 187). Three ablation strategies were applied for each LA model: pulmonary vein isolation (PVI), fibrosis-based ablation (FIBRO) and a rotor-based ablation (ROTOR). The DL model was trained to predict the success of each RFCA strategy for each LA model. Three feature attribution (FA) map methods were then used to investigate interpretability of the DL model: GradCAM, Occlusions and LIME. Results: The developed DL model had an AUC (area under the receiver operating characteristic curve) of 0.78 ± 0.04 for predicting the success of the PVI strategy, 0.92 ± 0.02 for FIBRO and 0.77 ± 0.02 for ROTOR. GradCAM had the highest percentage of informative regions in the FA maps (62% for FIBRO and 71% for ROTOR) that coincided with the successful RFCA lesions known from the 2D LA simulations, but unseen by the DL model. Moreover, GradCAM had the smallest coincidence of informative regions of the FA maps with non-arrhythmogenic regions (25% for FIBRO and 27% for ROTOR). Conclusion: The most informative regions of the FA maps coincided with pro-arrhythmogenic regions, suggesting that the DL model leveraged structural features of MRI images to identify such regions and make its prediction. In the future, this technique could provide a clinician with a trustworthy decision support tool.

Real-time speech MRI datasets with corresponding articulator ground-truth segmentations.

The use of real-time magnetic resonance imaging (rt-MRI) of speech is increasing in clinical practice and speech science research. Analysis of such images often requires segmentation of articulators and the vocal tract, and the community is turning to deep-learning-based methods to perform this segmentation. While there are publicly available rt-MRI datasets of speech, these do not include ground-truth (GT) segmentations, a key requirement for the development of deep-learning-based segmentation methods. To begin to address this barrier, this work presents rt-MRI speech datasets of five healthy adult volunteers with corresponding GT segmentations and velopharyngeal closure patterns. The images were acquired using standard clinical MRI scanners, coils and sequences to facilitate acquisition of similar images in other centres. The datasets include manually created GT segmentations of six anatomical features including the tongue, soft palate and vocal tract. In addition, this work makes code and instructions to implement a current state-of-the-art deep-learning-based method to segment rt-MRI speech datasets publicly available, thus providing the community and others with a starting point for developing such methods.

Hemodynamic Determinants of Elevated Blood Pressure and Hypertension in the Middle to Older-Age UK Population: A UK Biobank Imaging Study.

BACKGROUND: Increased systemic vascular resistance and, in older people, reduced aortic distensibility, are thought to be the hemodynamic determinants of primary hypertension but cardiac output could also be important. We examined the hemodynamics of elevated blood pressure and hypertension in the middle to older-aged UK population participating in the UK Biobank imaging studies. METHODS: Cardiac output, systemic vascular resistance, and aortic distensibility were measured from cardiac magnetic resonance imaging in 31 112 (distensibility in 21 178) participants (46.3% male, mean age±SD 63±7 years). Body composition including visceral adipose tissue volume and abdominal subcutaneous adipose tissue volume were measured in 19 645 participants. RESULTS: Participants with higher blood pressure had higher cardiac output (higher by 17.9±26.6% in hypertensive compared with those with optimal blood pressure) and higher systemic vascular resistance (higher by 11.4±27.9% in hypertensive compared with those with optimal blood pressure). These differences were little changed after adjustment for body size and adiposity. The contribution of cardiac output relative to systemic vascular resistance was more marked in younger compared with older subjects. Aortic distensibility decreased with age and was lower in participants with higher compared with lower blood pressure but with a greater difference in younger compared with older subjects. CONCLUSIONS: In the middle to older-aged UK population, cardiac output plays an important role in contributing to elevated mean arterial blood pressure, particularly in younger compared with older subjects. Reduced aortic distensibility contributes to a rise in pulse pressure and systolic blood pressure at all ages.

Healthcare Experiences of African American Women with the Fragile X Premutation.

This study aims to understand the healthcare experiences of African American women with a fragile X premutation (PM). PM carriers are at risk for fragile X-associated conditions, including primary ovarian insufficiency (FXPOI) and neuropsychiatric disorders (FXAND). There is no racial/ethnic association with carrying a PM, but African American women historically experience barriers receiving quality healthcare in the USA. Obstacles to care may increase mental health conditions like anxiety and depression. Eight African American women with a PM were interviewed to explore disparities in receiving healthcare and to learn about psychosocial experiences during and after their diagnoses. Interviews were transcribed verbatim and independently coded by two researchers. A deductive-inductive approach was used, followed by thematic analysis to determine prominent themes. The average participant age was 52.3 ± 8.60 years, with a mean age at premutation diagnosis of 31 ± 5.95 years. Seven participants had children with FXS. Themes from interviews included healthcare experiences, family dynamics, and emotional/mental health after their diagnosis. Participants reported concerns about not being taken seriously by providers and mistrust of the medical institutions. Within families, participants reported denial, insensitivity, and isolation. Participants reported a high incidence of anxiety and depression. Both are symptoms of FXAND and stresses of systemic racism and sexism. The reported family dynamics around the news of a genetic diagnosis stand apart from other racial cohorts in fragile X research: interventions like family counseling sessions and inclusive support opportunities from national organizations could ease the impacts of a PM for African American women.

Uncertainty aware training to improve deep learning model calibration for classification of cardiac MR images.

Quantifying uncertainty of predictions has been identified as one way to develop more trustworthy artificial intelligence (AI) models beyond conventional reporting of performance metrics. When considering their role in a clinical decision support setting, AI classification models should ideally avoid confident wrong predictions and maximise the confidence of correct predictions. Models that do this are said to be well calibrated with regard to confidence. However, relatively little attention has been paid to how to improve calibration when training these models, i.e. to make the training strategy uncertainty-aware. In this work we: (i) evaluate three novel uncertainty-aware training strategies with regard to a range of accuracy and calibration performance measures, comparing against two state-of-the-art approaches, (ii) quantify the data (aleatoric) and model (epistemic) uncertainty of all models and (iii) evaluate the impact of using a model calibration measure for model selection in uncertainty-aware training, in contrast to the normal accuracy-based measures. We perform our analysis using two different clinical applications: cardiac resynchronisation therapy (CRT) response prediction and coronary artery disease (CAD) diagnosis from cardiac magnetic resonance (CMR) images. The best-performing model in terms of both classification accuracy and the most common calibration measure, expected calibration error (ECE) was the Confidence Weight method, a novel approach that weights the loss of samples to explicitly penalise confident incorrect predictions. The method reduced the ECE by 17% for CRT response prediction and by 22% for CAD diagnosis when compared to a baseline classifier in which no uncertainty-aware strategy was included. In both applications, as well as reducing the ECE there was a slight increase in accuracy from 69% to 70% and 70% to 72% for CRT response prediction and CAD diagnosis respectively. However, our analysis showed a lack of consistency in terms of optimal models when using different calibration measures. This indicates the need for careful consideration of performance metrics when training and selecting models for complex high risk applications in healthcare.

An artificial intelligence tool for automated analysis of large-scale unstructured clinical cine cardiac magnetic resonance databases.

Aims: Artificial intelligence (AI) techniques have been proposed for automating analysis of short-axis (SAX) cine cardiac magnetic resonance (CMR), but no CMR analysis tool exists to automatically analyse large (unstructured) clinical CMR datasets. We develop and validate a robust AI tool for start-to-end automatic quantification of cardiac function from SAX cine CMR in large clinical databases. Methods and results: Our pipeline for processing and analysing CMR databases includes automated steps to identify the correct data, robust image pre-processing, an AI algorithm for biventricular segmentation of SAX CMR and estimation of functional biomarkers, and automated post-analysis quality control to detect and correct errors. The segmentation algorithm was trained on 2793 CMR scans from two NHS hospitals and validated on additional cases from this dataset (n = 414) and five external datasets (n = 6888), including scans of patients with a range of diseases acquired at 12 different centres using CMR scanners from all major vendors. Median absolute errors in cardiac biomarkers were within the range of inter-observer variability: <8.4 mL (left ventricle volume), <9.2 mL (right ventricle volume), <13.3 g (left ventricular mass), and <5.9% (ejection fraction) across all datasets. Stratification of cases according to phenotypes of cardiac disease and scanner vendors showed good performance across all groups. Conclusion: We show that our proposed tool, which combines image pre-processing steps, a domain-generalizable AI algorithm trained on a large-scale multi-domain CMR dataset and quality control steps, allows robust analysis of (clinical or research) databases from multiple centres, vendors, and cardiac diseases. This enables translation of our tool for use in fully automated processing of large multi-centre databases.

Environmental and genetic predictors of human cardiovascular ageing.

Cardiovascular ageing is a process that begins early in life and leads to a progressive change in structure and decline in function due to accumulated damage across diverse cell types, tissues and organs contributing to multi-morbidity. Damaging biophysical, metabolic and immunological factors exceed endogenous repair mechanisms resulting in a pro-fibrotic state, cellular senescence and end-organ damage, however the genetic architecture of cardiovascular ageing is not known. Here we use machine learning approaches to quantify cardiovascular age from image-derived traits of vascular function, cardiac motion and myocardial fibrosis, as well as conduction traits from electrocardiograms, in 39,559 participants of UK Biobank. Cardiovascular ageing is found to be significantly associated with common or rare variants in genes regulating sarcomere homeostasis, myocardial immunomodulation, and tissue responses to biophysical stress. Ageing is accelerated by cardiometabolic risk factors and we also identify prescribed medications that are potential modifiers of ageing. Through large-scale modelling of ageing across multiple traits our results reveal insights into the mechanisms driving premature cardiovascular ageing and reveal potential molecular targets to attenuate age-related processes.

A Topological Loss Function for Deep-Learning Based Image Segmentation Using Persistent Homology.

We introduce a method for training neural networks to perform image or volume segmentation in which prior knowledge about the topology of the segmented object can be explicitly provided and then incorporated into the training process. By using the differentiable properties of persistent homology, a concept used in topological data analysis, we can specify the desired topology of segmented objects in terms of their Betti numbers and then drive the proposed segmentations to contain the specified topological features. Importantly this process does not require any ground-truth labels, just prior knowledge of the topology of the structure being segmented. We demonstrate our approach in four experiments: one on MNIST image denoising and digit recognition, one on left ventricular myocardium segmentation from magnetic resonance imaging data from the UK Biobank, one on the ACDC public challenge dataset and one on placenta segmentation from 3-D ultrasound. We find that embedding explicit prior knowledge in neural network segmentation tasks is most beneficial when the segmentation task is especially challenging and that it can be used in either a semi-supervised or post-processing context to extract a useful training gradient from images without pixelwise labels.

Aortic Distensibility Measured by Automated Analysis of Magnetic Resonance Imaging Predicts Adverse Cardiovascular Events in UK Biobank.

Background Automated analysis of cardiovascular magnetic resonance images provides the potential to assess aortic distensibility in large populations. The aim of this study was to compare the prediction of cardiovascular events by automated cardiovascular magnetic resonance with those of other simple measures of aortic stiffness suitable for population screening. Methods and Results Aortic distensibility was measured from automated segmentation of aortic cine cardiovascular magnetic resonance using artificial intelligence in 8435 participants. The associations of distensibility, brachial pulse pressure, and stiffness index (obtained by finger photoplethysmography) with conventional risk factors was examined by multivariable regression and incident cardiovascular events by Cox proportional-hazards regression. Mean (±SD) distensibility values for men and women were 1.77±1.15 and 2.10±1.45 (P<0.0001) 10-3 mm Hg-1, respectively. There was a good correlation between automatically and manually obtained systolic and diastolic aortic areas (r=0.980 and r=0.985, respectively). In regression analysis, distensibility associated with age, mean arterial pressure, heart rate, weight, and plasma glucose but not male sex, cholesterol or current smoking. During an average follow-up of 2.8±1.3 years, 86 participants experienced cardiovascular events 6 of whom died. Higher distensibility was associated with reduced risk of cardiovascular events (adjusted hazard ratio [HR], 0.61 per log unit of distensibility; P=0.016). There was no evidence of an association between pulse pressure (adjusted HR 1.00; P=0.715) or stiffness index (adjusted HR, 1.02; P=0.535) and risk of cardiovascular events. Conclusions Automated cardiovascular magnetic resonance-derived aortic distensibility may be incorporated into routine clinical imaging. It shows a similar association to cardiovascular risk factors as other measures of arterial stiffness and predicts new-onset cardiovascular events, making it a useful tool for the measurement of vascular aging and associated cardiovascular risk.