Deep learning models for automatic tumor segmentation and total tumor volume assessment in patients with colorectal liver metastases.

BACKGROUND: We developed models for tumor segmentation to automate the assessment of total tumor volume (TTV) in patients with colorectal liver metastases (CRLM). METHODS: In this prospective cohort study, pre- and post-systemic treatment computed tomography (CT) scans of 259 patients with initially unresectable CRLM of the CAIRO5 trial (NCT02162563) were included. In total, 595 CT scans comprising 8,959 CRLM were divided into training (73%), validation (6.5%), and test sets (21%). Deep learning models were trained with ground truth segmentations of the liver and CRLM. TTV was calculated based on the CRLM segmentations. An external validation cohort was included, comprising 72 preoperative CT scans of patients with 112 resectable CRLM. Image segmentation evaluation metrics and intraclass correlation coefficient (ICC) were calculated. RESULTS: In the test set (122 CT scans), the autosegmentation models showed a global Dice similarity coefficient (DSC) of 0.96 (liver) and 0.86 (CRLM). The corresponding median per-case DSC was 0.96 (interquartile range [IQR] 0.95-0.96) and 0.80 (IQR 0.67-0.87). For tumor segmentation, the intersection-over-union, precision, and recall were 0.75, 0.89, and 0.84, respectively. An excellent agreement was observed between the reference and automatically computed TTV for the test set (ICC 0.98) and external validation cohort (ICC 0.98). In the external validation, the global DSC was 0.82 and the median per-case DSC was 0.60 (IQR 0.29-0.76) for tumor segmentation. CONCLUSIONS: Deep learning autosegmentation models were able to segment the liver and CRLM automatically and accurately in patients with initially unresectable CRLM, enabling automatic TTV assessment in such patients. RELEVANCE STATEMENT: Automatic segmentation enables the assessment of total tumor volume in patients with colorectal liver metastases, with a high potential of decreasing radiologist's workload and increasing accuracy and consistency. KEY POINTS: • Tumor response evaluation is time-consuming, manually performed, and ignores total tumor volume. • Automatic models can accurately segment tumors in patients with colorectal liver metastases. • Total tumor volume can be accurately calculated based on automatic segmentations.

Arrhythmia risk stratification of patients after myocardial infarction using personalized heart models.

Sudden cardiac death (SCD) from arrhythmias is a leading cause of mortality. For patients at high SCD risk, prophylactic insertion of implantable cardioverter defibrillators (ICDs) reduces mortality. Current approaches to identify patients at risk for arrhythmia are, however, of low sensitivity and specificity, which results in a low rate of appropriate ICD therapy. Here, we develop a personalized approach to assess SCD risk in post-infarction patients based on cardiac imaging and computational modelling. We construct personalized three-dimensional computer models of post-infarction hearts from patients' clinical magnetic resonance imaging data and assess the propensity of each model to develop arrhythmia. In a proof-of-concept retrospective study, the virtual heart test significantly outperformed several existing clinical metrics in predicting future arrhythmic events. The robust and non-invasive personalized virtual heart risk assessment may have the potential to prevent SCD and avoid unnecessary ICD implantations.

Correction: Virtual Electrophysiological Study of Atrial Fibrillation in Fibrotic Remodeling.

[This corrects the article DOI: 10.1371/journal.pone.0117110.].

Myocardial Infarct Segmentation From Magnetic Resonance Images for Personalized Modeling of Cardiac Electrophysiology.

Accurate representation of myocardial infarct geometry is crucial to patient-specific computational modeling of the heart in ischemic cardiomyopathy. We have developed a methodology for segmentation of left ventricular (LV) infarct from clinically acquired, two-dimensional (2D), late-gadolinium enhanced cardiac magnetic resonance (LGE-CMR) images, for personalized modeling of ventricular electrophysiology. The infarct segmentation was expressed as a continuous min-cut optimization problem, which was solved using its dual formulation, the continuous max-flow (CMF). The optimization objective comprised of a smoothness term, and a data term that quantified the similarity between image intensity histograms of segmented regions and those of a set of training images. A manual segmentation of the LV myocardium was used to initialize and constrain the developed method. The three-dimensional geometry of infarct was reconstructed from its segmentation using an implicit, shape-based interpolation method. The proposed methodology was extensively evaluated using metrics based on geometry, and outcomes of individualized electrophysiological simulations of cardiac dys(function). Several existing LV infarct segmentation approaches were implemented, and compared with the proposed method. Our results demonstrated that the CMF method was more accurate than the existing approaches in reproducing expert manual LV infarct segmentations, and in electrophysiological simulations. The infarct segmentation method we have developed and comprehensively evaluated in this study constitutes an important step in advancing clinical applications of personalized simulations of cardiac electrophysiology.

Image-based Reconstruction of 3D Myocardial Infarct Geometry for Patient Specific Applications.

Accurate reconstruction of the three-dimensional (3D) geometry of a myocardial infarct from two-dimensional (2D) multi-slice image sequences has important applications in the clinical evaluation and treatment of patients with ischemic cardiomyopathy. However, this reconstruction is challenging because the resolution of common clinical scans used to acquire infarct structure, such as short-axis, late-gadolinium enhanced cardiac magnetic resonance (LGE-CMR) images, is low, especially in the out-of-plane direction. In this study, we propose a novel technique to reconstruct the 3D infarct geometry from low resolution clinical images. Our methodology is based on a function called logarithm of odds (LogOdds), which allows the broader class of linear combinations in the LogOdds vector space as opposed to being limited to only a convex combination in the binary label space. To assess the efficacy of the method, we used high-resolution LGE-CMR images of 36 human hearts in vivo, and 3 canine hearts ex vivo. The infarct was manually segmented in each slice of the acquired images, and the manually segmented data were downsampled to clinical resolution. The developed method was then applied to the downsampled image slices, and the resulting reconstructions were compared with the manually segmented data. Several existing reconstruction techniques were also implemented, and compared with the proposed method. The results show that the LogOdds method significantly outperforms all the other tested methods in terms of region overlap.

Image-based reconstruction of three-dimensional myocardial infarct geometry for patient-specific modeling of cardiac electrophysiology.

PURPOSE: Accurate three-dimensional (3D) reconstruction of myocardial infarct geometry is crucial to patient-specific modeling of the heart aimed at providing therapeutic guidance in ischemic cardiomyopathy. However, myocardial infarct imaging is clinically performed using two-dimensional (2D) late-gadolinium enhanced cardiac magnetic resonance (LGE-CMR) techniques, and a method to build accurate 3D infarct reconstructions from the 2D LGE-CMR images has been lacking. The purpose of this study was to address this need. METHODS: The authors developed a novel methodology to reconstruct 3D infarct geometry from segmented low-resolution (Lo-res) clinical LGE-CMR images. Their methodology employed the so-called logarithm of odds (LogOdds) function to implicitly represent the shape of the infarct in segmented image slices as LogOdds maps. These 2D maps were then interpolated into a 3D image, and the result transformed via the inverse of LogOdds to a binary image representing the 3D infarct geometry. To assess the efficacy of this method, the authors utilized 39 high-resolution (Hi-res) LGE-CMR images, including 36 in vivo acquisitions of human subjects with prior myocardial infarction and 3 ex vivo scans of canine hearts following coronary ligation to induce infarction. The infarct was manually segmented by trained experts in each slice of the Hi-res images, and the segmented data were downsampled to typical clinical resolution. The proposed method was then used to reconstruct 3D infarct geometry from the downsampled images, and the resulting reconstructions were compared with the manually segmented data. The method was extensively evaluated using metrics based on geometry as well as results of electrophysiological simulations of cardiac sinus rhythm and ventricular tachycardia in individual hearts. Several alternative reconstruction techniques were also implemented and compared with the proposed method. RESULTS: The accuracy of the LogOdds method in reconstructing 3D infarct geometry, as measured by the Dice similarity coefficient, was 82.10% ± 6.58%, a significantly higher value than those of the alternative reconstruction methods. Among outcomes of electrophysiological simulations with infarct reconstructions generated by various methods, the simulation results corresponding to the LogOdds method showed the smallest deviation from those corresponding to the manual reconstructions, as measured by metrics based on both activation maps and pseudo-ECGs. CONCLUSIONS: The authors have developed a novel method for reconstructing 3D infarct geometry from segmented slices of Lo-res clinical 2D LGE-CMR images. This method outperformed alternative approaches in reproducing expert manual 3D reconstructions and in electrophysiological simulations.

Virtual electrophysiological study of atrial fibrillation in fibrotic remodeling.

Research has indicated that atrial fibrillation (AF) ablation failure is related to the presence of atrial fibrosis. However it remains unclear whether this information can be successfully used in predicting the optimal ablation targets for AF termination. We aimed to provide a proof-of-concept that patient-specific virtual electrophysiological study that combines i) atrial structure and fibrosis distribution from clinical MRI and ii) modeling of atrial electrophysiology, could be used to predict: (1) how fibrosis distribution determines the locations from which paced beats degrade into AF; (2) the dynamic behavior of persistent AF rotors; and (3) the optimal ablation targets in each patient. Four MRI-based patient-specific models of fibrotic left atria were generated, ranging in fibrosis amount. Virtual electrophysiological studies were performed in these models, and where AF was inducible, the dynamics of AF were used to determine the ablation locations that render AF non-inducible. In 2 of the 4 models patient-specific models AF was induced; in these models the distance between a given pacing location and the closest fibrotic region determined whether AF was inducible from that particular location, with only the mid-range distances resulting in arrhythmia. Phase singularities of persistent rotors were found to move within restricted regions of tissue, which were independent of the pacing location from which AF was induced. Electrophysiological sensitivity analysis demonstrated that these regions changed little with variations in electrophysiological parameters. Patient-specific distribution of fibrosis was thus found to be a critical component of AF initiation and maintenance. When the restricted regions encompassing the meander of the persistent phase singularities were modeled as ablation lesions, AF could no longer be induced. The study demonstrates that a patient-specific modeling approach to identify non-invasively AF ablation targets prior to the clinical procedure is feasible.

Myocardial infarct segmentation and reconstruction from 2D late-gadolinium enhanced magnetic resonance images.

In this paper, we propose a convex optimization-based algorithm for segmenting myocardial infarct from clinical 2D late-gadolinium enhanced magnetic resonance (LGE-MR) images. Previously segmented left ventricular (LV) myocardium was used to define a region of interest for the infarct segmentation. The infarct segmentation problem was formulated as a continuous min-cut problem, which was solved using its dual formulation, the continuous max-flow (CMF). Bhattacharyya intensity distribution matching was used as the data term, where the prior intensity distributions were computed based on a training data set LGE-MR images from seven patients. The algorithm was parallelized and implemented in a graphics processing unit for reduced computation time. Three-dimensional (3D) volumes of the infarcts were then reconstructed using an interpolation technique we developed based on logarithm of odds. The algorithm was validated using LGE-MR images from 47 patients (309 slices) by comparing computed 2D segmentations and 3D reconstructions to manually generated ones. In addition, the developed algorithm was compared to several previously reported segmentation techniques. The CMF algorithm outperformed the previously reported methods in terms of Dice similarity coefficient.

Image-based left ventricular shape analysis for sudden cardiac death risk stratification.

BACKGROUND: Low left ventricular ejection fraction (LVEF), the main criterion used in the current clinical practice to stratify sudden cardiac death (SCD) risk, has low sensitivity and specificity. OBJECTIVE: To uncover indices of left ventricular (LV) shape that differ between patients with a high risk of SCD and those with a low risk. METHODS: By using clinical cardiac magnetic resonance imaging and computational anatomy tools, a novel computational framework to compare 3-dimensional LV endocardial surface curvedness, wall thickness, and relative wall thickness between patient groups was implemented. The framework was applied to cardiac magnetic resonance data of 61 patients with ischemic cardiomyopathy who were selected for prophylactic implantable cardioverter-defibrillator treatment on the basis of reduced LVEF. The patients were classified by outcome: group 0 had no events; group 1, arrhythmic events; and group 2, heart failure events. Segmental differences in LV shape were assessed. RESULTS: Global LV volumes and mass were similar among groups. Compared with patients with no events, patients in groups 1 and 2 had lower mean shape metrics in all coronary artery regions, with statistical significance in 9 comparisons, reflecting wall thinning and stretching/flattening. CONCLUSION: In patients with ischemic cardiomyopathy and low LVEF, there exist quantifiable differences in 3-dimensional endocardial surface curvedness, LV wall thickness, and LV relative wall thickness between those with no clinical events and those with arrhythmic or heart failure outcomes, reflecting adverse LV remodeling. This retrospective study is a proof of concept to demonstrate that regional LV remodeling indices have the potential to improve the personalized risk assessment for SCD.

Mechanistic inquiry into the role of tissue remodeling in fibrotic lesions in human atrial fibrillation.

Atrial fibrillation (AF), the most common arrhythmia in humans, is initiated when triggered activity from the pulmonary veins propagates into atrial tissue and degrades into reentrant activity. Although experimental and clinical findings show a correlation between atrial fibrosis and AF, the causal relationship between the two remains elusive. This study used an array of 3D computational models with different representations of fibrosis based on a patient-specific atrial geometry with accurate fibrotic distribution to determine the mechanisms by which fibrosis underlies the degradation of a pulmonary vein ectopic beat into AF. Fibrotic lesions in models were represented with combinations of: gap junction remodeling; collagen deposition; and myofibroblast proliferation with electrotonic or paracrine effects on neighboring myocytes. The study found that the occurrence of gap junction remodeling and the subsequent conduction slowing in the fibrotic lesions was a necessary but not sufficient condition for AF development, whereas myofibroblast proliferation and the subsequent electrophysiological effect on neighboring myocytes within the fibrotic lesions was the sufficient condition necessary for reentry formation. Collagen did not alter the arrhythmogenic outcome resulting from the other fibrosis components. Reentrant circuits formed throughout the noncontiguous fibrotic lesions, without anchoring to a specific fibrotic lesion.

Placement of implantable cardioverter-defibrillators in paediatric and congenital heart defect patients: a pipeline for model generation and simulation prediction of optimal configurations.

There is currently no reliable way of predicting the optimal implantable cardioverter-defibrillator (ICD) placement in paediatric and congenital heart defect (CHD) patients. This study aimed to: (1) develop a new image processing pipeline for constructing patient-specific heart-torso models from clinical magnetic resonance images (MRIs); (2) use the pipeline to determine the optimal ICD configuration in a paediatric tricuspid valve atresia patient; (3) establish whether the widely used criterion of shock-induced extracellular potential (Φe) gradients ≥5 V cm(-1) in ≥95% of ventricular volume predicts defibrillation success. A biophysically detailed heart-torso model was generated from patient MRIs. Because transvenous access was impossible, three subcutaneous and three epicardial lead placement sites were identified along with five ICD scan locations. Ventricular fibrillation was induced, and defibrillation shocks were applied from 11 ICD configurations to determine defibrillation thresholds (DFTs). Two configurations with epicardial leads resulted in the lowest DFTs overall and were thus considered optimal. Three configurations shared the lowest DFT among subcutaneous lead ICDs. The Φe gradient criterion was an inadequate predictor of defibrillation success, as defibrillation failed in numerous instances even when 100% of the myocardium experienced such gradients. In conclusion, we have developed a new image processing pipeline and applied it to a CHD patient to construct the first active heart-torso model from clinical MRIs.

Feasibility of image-based simulation to estimate ablation target in human ventricular arrhythmia.

BACKGROUND: Previous studies suggest that magnetic resonance imaging with late gadolinium enhancement (LGE) may identify slowly conducting tissues in scar-related ventricular tachycardia (VT). OBJECTIVE: To test the feasibility of image-based simulation based on LGE to estimate ablation targets in VT. METHODS: We conducted a retrospective study in 13 patients who had preablation magnetic resonance imaging for scar-related VT ablation. We used image-based simulation to induce VT and estimate target regions according to the simulated VT circuit. The estimated target regions were coregistered with the LGE scar map and the ablation sites from the electroanatomical map in the standard ablation approach. RESULTS: In image-based simulation, VT was inducible in 12 (92.3%) patients. All VTs showed macroreentrant propagation patterns, and the narrowest width of estimated target region that an ablation line should span to prevent VT recurrence was 5.0 ± 3.4 mm. Of 11 patients who underwent ablation, the results of image-based simulation and the standard approach were consistent in 9 (82%) patients, where ablation within the estimated target region was associated with acute success (n = 8) and ablation outside the estimated target region was associated with failure (n = 1). In 1 (9%) case, the results of image-based simulation and the standard approach were inconsistent, where ablation outside the estimated target region was associated with acute success. CONCLUSIONS: The image-based simulation can be used to estimate potential ablation targets of scar-related VT. The image-based simulation may be a powerful noninvasive tool for preprocedural planning of ablation procedures to potentially reduce the procedure time and complication rates.

Patient-specific modeling of the heart: estimation of ventricular fiber orientations.

Patient-specific simulations of heart (dys)function aimed at personalizing cardiac therapy are hampered by the absence of in vivo imaging technology for clinically acquiring myocardial fiber orientations. The objective of this project was to develop a methodology to estimate cardiac fiber orientations from in vivo images of patient heart geometries. An accurate representation of ventricular geometry and fiber orientations was reconstructed, respectively, from high-resolution ex vivo structural magnetic resonance (MR) and diffusion tensor (DT) MR images of a normal human heart, referred to as the atlas. Ventricular geometry of a patient heart was extracted, via semiautomatic segmentation, from an in vivo computed tomography (CT) image. Using image transformation algorithms, the atlas ventricular geometry was deformed to match that of the patient. Finally, the deformation field was applied to the atlas fiber orientations to obtain an estimate of patient fiber orientations. The accuracy of the fiber estimates was assessed using six normal and three failing canine hearts. The mean absolute difference between inclination angles of acquired and estimated fiber orientations was 15.4 °. Computational simulations of ventricular activation maps and pseudo-ECGs in sinus rhythm and ventricular tachycardia indicated that there are no significant differences between estimated and acquired fiber orientations at a clinically observable level.The new insights obtained from the project will pave the way for the development of patient-specific models of the heart that can aid physicians in personalized diagnosis and decisions regarding electrophysiological interventions.

Computational cardiology: how computer simulations could be used to develop new therapies and advance existing ones.

This article reviews the latest developments in computational cardiology. It focuses on the contribution of cardiac modelling to the development of new therapies as well as the advancement of existing ones for cardiac arrhythmias and pump dysfunction. Reviewed are cardiac modelling efforts aimed at advancing and optimizing existent therapies for cardiac disease (defibrillation, ablation of ventricular tachycardia, and cardiac resynchronization therapy) and at suggesting novel treatments, including novel molecular targets, as well as efforts to use cardiac models in stratification of patients likely to benefit from a given therapy, and the use of models in diagnostic procedures.

Methodology for patient-specific modeling of atrial fibrosis as a substrate for atrial fibrillation.

Personalized computational cardiac models are emerging as an important tool for studying cardiac arrhythmia mechanisms, and have the potential to become powerful instruments for guiding clinical anti-arrhythmia therapy. In this article, we present the methodology for constructing a patient-specific model of atrial fibrosis as a substrate for atrial fibrillation. The model is constructed from high-resolution late gadolinium-enhanced magnetic resonance imaging (LGE-MRI) images acquired in vivo from a patient suffering from persistent atrial fibrillation, accurately capturing both the patient's atrial geometry and the distribution of the fibrotic regions in the atria. Atrial fiber orientation is estimated using a novel image-based method, and fibrosis is represented in the patient-specific fibrotic regions as incorporating collagenous septa, gap junction remodeling, and myofibroblast proliferation. A proof-of-concept simulation result of reentrant circuits underlying atrial fibrillation in the model of the patient's fibrotic atrium is presented to demonstrate the completion of methodology development.

Image-based estimation of ventricular fiber orientations for personalized modeling of cardiac electrophysiology.

Technological limitations pose a major challenge to acquisition of myocardial fiber orientations for patient-specific modeling of cardiac (dys)function and assessment of therapy. The objective of this project was to develop a methodology to estimate cardiac fiber orientations from in vivo images of patient heart geometries. An accurate representation of ventricular geometry and fiber orientations was reconstructed, respectively, from high-resolution ex vivo structural magnetic resonance (MR) and diffusion tensor (DT) MR images of a normal human heart, referred to as the atlas. Ventricular geometry of a patient heart was extracted, via semiautomatic segmentation, from an in vivo computed tomography (CT) image. Using image transformation algorithms, the atlas ventricular geometry was deformed to match that of the patient. Finally, the deformation field was applied to the atlas fiber orientations to obtain an estimate of patient fiber orientations. The accuracy of the fiber estimates was assessed using six normal and three failing canine hearts. The mean absolute difference between inclination angles of acquired and estimated fiber orientations was 15.4°. Computational simulations of ventricular activation maps and pseudo-ECGs in sinus rhythm and ventricular tachycardia indicated that there are no significant differences between estimated and acquired fiber orientations at a clinically observable level.

Left-ventricular shape analysis for predicting sudden cardiac death risk.

Implantation of cardioverter defibrillators is the most widely used primary preventive care for sudden cardiac death (SCD). Current clinical practice of using a left-ventricular ejection fraction threshold as the sole criterion for defibrillator insertion results in many unnecessary implantations. To address the need for alternative criteria, we seek three-dimensional shape metrics of the left ventricle derived from clinical cardiac magnetic resonance images that can predict SCD risk. The present study is a proof-of-concept, where we have combined image-processing and computational anatomy techniques to develop a processing pipeline to statistically compare localized left ventricular shape metrics between patient groups. We tested the methodology with data from a small cohort of patients, classified into two groups based on SCD risk. The results demonstrate that our approach is able to locate systematic wall thickness differences between the two groups.

Image-based estimation of ventricular fiber orientations for patient-specific simulations.

Patient-specific simulation of heart (dys)function aimed at personalizing cardiac therapy are hampered by the absence of in vivo imaging technology for clinically acquiring myocardial fiber orientations. In this research, we develop a methodology to predict ventricular fiber orientations of a patient heart, given the geometry of the heart and an atlas. We test the methodology by comparing the estimated fiber orientations with measured ones, and by quantifying the effect of the estimation error on outcomes of electrophysiological simulations, in normal and failing canine hearts. The new insights obtained from the project will pave the way for the development of patient-specific models of the heart that can aid physicians in personalized diagnosis and decisions regarding electrophysiological interventions.

Image-based models of cardiac structure in health and disease.

Computational approaches to investigating the electromechanics of healthy and diseased hearts are becoming essential for the comprehensive understanding of cardiac function. In this article, we first present a brief review of existing image-based computational models of cardiac structure. We then provide a detailed explanation of a processing pipeline which we have recently developed for constructing realistic computational models of the heart from high resolution structural and diffusion tensor (DT) magnetic resonance (MR) images acquired ex vivo. The presentation of the pipeline incorporates a review of the methodologies that can be used to reconstruct models of cardiac structure. In this pipeline, the structural image is segmented to reconstruct the ventricles, normal myocardium, and infarct. A finite element mesh is generated from the segmented structural image, and fiber orientations are assigned to the elements based on DTMR data. The methods were applied to construct seven different models of healthy and diseased hearts. These models contain millions of elements, with spatial resolutions in the order of hundreds of microns, providing unprecedented detail in the representation of cardiac structure for simulation studies.

Towards predictive modelling of the electrophysiology of the heart.

The simulation of cardiac electrical function is an example of a successful integrative multiscale modelling approach that is directly relevant to human disease. Today we stand at the threshold of a new era, in which anatomically detailed, tomographically reconstructed models are being developed that integrate from the ion channel to the electromechanical interactions in the intact heart. Such models hold high promise for interpretation of clinical and physiological measurements, for improving the basic understanding of the mechanisms of dysfunction in disease, such as arrhythmias, myocardial ischaemia and heart failure, and for the development and performance optimization of medical devices. The goal of this article is to present an overview of current state-of-art advances towards predictive computational modelling of the heart as developed recently by the authors of this article. We first outline the methodology for constructing electrophysiological models of the heart. We then provide three examples that demonstrate the use of these models, focusing specifically on the mechanisms for arrhythmogenesis and defibrillation in the heart. These include: (1) uncovering the role of ventricular structure in defibrillation; (2) examining the contribution of Purkinje fibres to the failure of the shock; and (3) using magnetic resonance imaging reconstructed heart models to investigate the re-entrant circuits formed in the presence of an infarct scar.