Detección de fibrosis en imágenes de cine de resonancia cardíaca mediante el uso de técnicas de inteligencia artificial / Fibrosis detection on MR-Cine images by using artificial intelligence techniques

RESUMEN Introducción: Las técnicas de inteligencia artificial han demostrado tener un gran potencial en el área de la cardiología, especialmente para identificar patrones imperceptibles para el ser humano. En este sentido, dichas técnicas parecen ser las adecuadas para identificar patrones en la textura del miocardio con el objetivo de identificar y cuantificar la fibrosis. Objetivos: Proponer un nuevo método de inteligencia artificial para identificar fibrosis en imágenes cine de resonancia cardíaca. Materiales y métodos: Se realizó un estudio retrospectivo observacional en 75 sujetos del Sanatorio San Carlos de Bariloche. El método propuesto analiza la textura del miocardio en las imágenes cine CMR (resonancia magnética cardíaca) mediante el uso de una red neuronal convolucional que determinar el daño local del tejido miocárdico. Resultados: Se observó una precisión del 89% para cuantificar el daño tisular local en el conjunto de datos de validación y de un 70% para el conjunto de prueba. Además, el análisis cualitativo realizado muestra una alta correlación espacial en la localización de la lesión. Conclusiones: El método propuesto permite identificar espacialmente la fibrosis únicamente utilizando la información de los estudios de cine de resonancia magnética nuclear, mostrando el potencial de la técnica propuesta para cuantificar la viabilidad miocárdica en un futuro o estudiar la etiología de las lesiones.

ABSTRACT Background: Artificial intelligence techniques have demonstrated great potential in cardiology, especially to detect imperceptible patterns for the human eye. In this sense, these techniques seem to be adequate to identify patterns in the myocardial texture which could lead to characterize and quantify fibrosis. Purpose: The aim of this study was to postulate a new artificial intelligence method to identify fibrosis in cine cardiac magnetic resonance (CMR) imaging. Methods: A retrospective observational study was carried out in a population of 75 subjects from a clinical center of San Carlos de Bariloche. The proposed method analyzes the myocardial texture in cine CMR images using a convolutional neural network to determine local myocardial tissue damage. Results: An accuracy of 89% for quantifying local tissue damage was observed for the validation data set and 70% for the test set. In addition, the qualitative analysis showed a high spatial correlation in lesion location. Conclusions: The postulated method enables to spatially identify fibrosis using only the information from cine nuclear magnetic resonance studies, demonstrating the potential of this technique to quantify myocardial viability in the future or to study the etiology of lesions.

Cuantificación automática de los volúmenes y función de ambos ventrículos en resonancia cardíaca. Propuesta y evaluación de un método de inteligencia artificial / Automatic Quantification of Volumes and Biventricular Function in Cardiac Resonance. Validation of a New Artificial Intelligence Approach

RESUMEN Introducción: Las técnicas de inteligencia artificial han demostrado tener un gran potencial en el área de la cardiología, especialmente para cuantificar la función cardíaca de ambos ventrículos, volumen, masa y fracción de eyección (FE). Sin embargo, su aplicación en la clínica no es directa, entre otros motivos por la poca reproducibilidad frente a casos de la práctica diaria. Objetivos: Propuesta y evaluación de una nueva herramienta de inteligencia artificial para cuantificar la función cardíaca de ambos ventrículos (volumen, masa y FE). Estudiar su robustez para su uso en la clínica y analizar los tiempos de cómputo respecto a los métodos convencionales. Materiales y métodos: Se analizaron en total 189 pacientes, 89 de un centro regional y 100 de un centro público. El método propuesto utiliza dos redes convolucionales incorporando información anatómica del corazón para reducir los errores de clasificación. Resultados: Se observa una alta concordancia (coeficiente de Pearson) entre la cuantificación manual y la propuesta para cuantificar la función cardíaca (0,98, 0,92, 0,96 y 0,8 para los volúmenes y para la FE de ambos ventrículos) en tiempos cercanos a los 5 seg. por estudio. Conclusiones: El método propuesto permite cuantificar los volúmenes y función de ambos ventrículos en segundos con una precisión comparable a la de un especialista.

ABSTRACT Background: Artificial intelligence techniques have shown great potential in cardiology, especially in quantifying cardiac biventricular function, volume, mass, and ejection fraction (EF). However, its use in clinical practice is not straightforward due to its poor reproducibility with cases from daily practice, among other reasons. Objectives: To validate a new artificial intelligence tool in order to quantify the cardiac biventricular function (volume, mass, and EF). To analyze its robustness in the clinical area, and the computational times compared with conventional methods. Methods: A total of 189 patients were analyzed: 89 from a regional center and 100 from a public center. The method proposes two convolutional networks that include anatomical information of the heart to reduce classification errors. Results: A high concordance (Pearson coefficient) was observed between manual quantification and the proposed quantification of cardiac function (0.98, 0.92, 0.96 and 0.8 for volumes and biventricular EF) in about 5 seconds per study. Conclusions: This method quantifies biventricular function and volumes in seconds with an accuracy equivalent to that of a specialist.

Left ventricle segmentation using a Bayesian approach with distance dependent shape priors.

We propose a method for segmentation of the left ventricle in magnetic resonance cardiac images. The framework consists of an initial Bayesian segmentation of the central slice of the volume. This segmentation is used to locate a shape prior for the LV myocardial tissue. This shape prior is determined using the fact that the myocardium is approximately annular as seen in the short-axis. Then a second Bayesian segmentation is performed to obtain the final result. This procedure is repeated for the rest of the slices. An extrapolation of the area of the LV is used to determine a stopping criterion. The method was evaluated on the databases of the Cardiac Atlas project. Our results demonstrate a suitable accuracy for myocardial segmentation (≈0.8 Dice's coefficient). For the endocardium and the epicardium the Dice's coefficients are 0.94 and 0.9 respectively. The accuracy was also evaluated in terms of the Hausdorff distance and the average distance. For the myocardium we obtain 8 mm and 2 mm respectively. Our results demonstrate the capability and merits of the proposed method to estimate the structure of the LV. The method requires minimal user input and generates results with quality comparable to more complex approaches. This paper suggests a new efficient approach for automatic LV quantification based on a Bayesian technique with shape priors with errors comparable to state-of-the-art techniques.

Automatic quantification of the LV function and mass: A deep learning approach for cardiovascular MRI.

OBJECTIVE: This paper proposes a novel approach for automatic left ventricle (LV) quantification using convolutional neural networks (CNN). METHODS: The general framework consists of one CNN for detecting the LV, and another for tissue classification. Also, three new deep learning architectures were proposed for LV quantification. These new CNNs introduce the ideas of sparsity and depthwise separable convolution into the U-net architecture, as well as, a residual learning strategy level-to-level. To this end, we extend the classical U-net architecture and use the generalized Jaccard distance as optimization objective function. RESULTS: The CNNs were trained and evaluated with 140 patients from two public cardiovascular magnetic resonance datasets (Sunnybrook and Cardiac Atlas Project) by using a 5-fold cross-validation strategy. Our results demonstrate a suitable accuracy for myocardial segmentation ( ∼ 0.9 Dice's coefficient), and a strong correlation with the most relevant physiological measures: 0.99 for end-diastolic and end-systolic volume, 0.97 for the left myocardial mass, 0.95 for the ejection fraction and 0.93 for the stroke volume and cardiac output. CONCLUSION: Our simulation and clinical evaluation results demonstrate the capability and merits of the proposed CNN to estimate different structural and functional features such as LV mass and EF which are commonly used for both diagnosis and treatment of different pathologies. SIGNIFICANCE: This paper suggests a new approach for automatic LV quantification based on deep learning where errors are comparable to the inter- and intra-operator ranges for manual contouring.

Influence of ultrasound speckle tracking strategies for motion and strain estimation.

Speckle Tracking is one of the most prominent techniques used to estimate the regional movement of the heart based on ultrasound acquisitions. Many different approaches have been proposed, proving their suitability to obtain quantitative and qualitative information regarding myocardial deformation, motion and function assessment. New proposals to improve the basic algorithm usually focus on one of these three steps: (1) the similarity measure between images and the speckle model; (2) the transformation model, i.e. the type of motion considered between images; (3) the optimization strategies, such as the use of different optimization techniques in the transformation step or the inclusion of structural information. While many contributions have shown their good performance independently, it is not always clear how they perform when integrated in a whole pipeline. Every step will have a degree of influence over the following and hence over the final result. Thus, a Speckle Tracking pipeline must be analyzed as a whole when developing novel methods, since improvements in a particular step might be undermined by the choices taken in further steps. This work presents two main contributions: (1) We provide a complete analysis of the influence of the different steps in a Speckle Tracking pipeline over the motion and strain estimation accuracy. (2) The study proposes a methodology for the analysis of Speckle Tracking systems specifically designed to provide an easy and systematic way to include other strategies. We close the analysis with some conclusions and recommendations that can be used as an orientation of the degree of influence of the models for speckle, the transformation models, interpolation schemes and optimization strategies over the estimation of motion features. They can be further use to evaluate and design new strategy into a Speckle Tracking system.

A maximum likelihood approach to diffeomorphic speckle tracking for 3D strain estimation in echocardiography.

The strain and strain-rate measures are commonly used for the analysis and assessment of regional myocardial function. In echocardiography (EC), the strain analysis became possible using Tissue Doppler Imaging (TDI). Unfortunately, this modality shows an important limitation: the angle between the myocardial movement and the ultrasound beam should be small to provide reliable measures. This constraint makes it difficult to provide strain measures of the entire myocardium. Alternative non-Doppler techniques such as Speckle Tracking (ST) can provide strain measures without angle constraints. However, the spatial resolution and the noisy appearance of speckle still make the strain estimation a challenging task in EC. Several maximum likelihood approaches have been proposed to statistically characterize the behavior of speckle, which results in a better performance of speckle tracking. However, those models do not consider common transformations to achieve the final B-mode image (e.g. interpolation). This paper proposes a new maximum likelihood approach for speckle tracking which effectively characterizes speckle of the final B-mode image. Its formulation provides a diffeomorphic scheme than can be efficiently optimized with a second-order method. The novelty of the method is threefold: First, the statistical characterization of speckle generalizes conventional speckle models (Rayleigh, Nakagami and Gamma) to a more versatile model for real data. Second, the formulation includes local correlation to increase the efficiency of frame-to-frame speckle tracking. Third, a probabilistic myocardial tissue characterization is used to automatically identify more reliable myocardial motions. The accuracy and agreement assessment was evaluated on a set of 16 synthetic image sequences for three different scenarios: normal, acute ischemia and acute dyssynchrony. The proposed method was compared to six speckle tracking methods. Results revealed that the proposed method is the most accurate method to measure the motion and strain with an average median motion error of 0.42 mm and a median strain error of 2.0 ± 0.9%, 2.1 ± 1.3% and 7.1 ± 4.9% for circumferential, longitudinal and radial strain respectively. It also showed its capability to identify abnormal segments with reduced cardiac function and timing differences for the dyssynchrony cases. These results indicate that the proposed diffeomorphic speckle tracking method provides robust and accurate motion and strain estimation.

Fully automatic detection of salient features in 3-d transesophageal images.

Most automated segmentation approaches to the mitral valve and left ventricle in 3-D echocardiography require a manual initialization. In this article, we propose a fully automatic scheme to initialize a multicavity segmentation approach in 3-D transesophageal echocardiography by detecting the left ventricle long axis, the mitral valve and the aortic valve location. Our approach uses a probabilistic and structural tissue classification to find structures such as the mitral and aortic valves; the Hough transform for circles to find the center of the left ventricle; and multidimensional dynamic programming to find the best position for the left ventricle long axis. For accuracy and agreement assessment, the proposed method was evaluated in 19 patients with respect to manual landmarks and as initialization of a multicavity segmentation approach for the left ventricle, the right ventricle, the left atrium, the right atrium and the aorta. The segmentation results revealed no statistically significant differences between manual and automated initialization in a paired t-test (p > 0.05). Additionally, small biases between manual and automated initialization were detected in the Bland-Altman analysis (bias, variance) for the left ventricle (-0.04, 0.10); right ventricle (-0.07, 0.18); left atrium (-0.01, 0.03); right atrium (-0.04, 0.13); and aorta (-0.05, 0.14). These results indicate that the proposed approach provides robust and accurate detection to initialize a multicavity segmentation approach without any user interaction.