Coupled Machine Learning and Finite Element Analysis of Heart Left Ventricle in Patients with Cardiomyopathy

Abstract

Automatic diagnosis of dilated cardiomyopathy based on ultrasound images is a challenging task,because pixel intensity levels in the images are not related to the physical properties of the tissue,as well as there is poor signal-to-noise ratio in images and weak echoes. Therefore,this study develops an automated methodology for 3D reconstruction and analysis of heart ventricle (LV) in patients with cardiomyopathy that can be divided into three main steps: (1) machine learning algorithms to extract LV and relevant parameters,(2) geometrical algorithms to reconstruct 3- dimensional model and (3) finite element method to analyse the mechanical response of the left ventricle under different loading conditions. Collected dataset consisted of 1809 images with apical view and 53 images with an M-mode view from cardiomyopathy patients at clinical facilities in the United Kingdom and Serbia. Methodology for analysis of apical view was based on segmentation of LV using U-net convolutional neural network [1] after which the segmented region was bounded with rectangle,where the longer side corresponds to left ventricular length (LVL). Regarding the M-mode view,traditional algorithms such as adaptive histogram equalization,template matching,Canny edge detection,and thresholding are used to extract internal dimension (LVID),posterior wall thickness (LVPW),and interventricular septum thickness (IVS),due to the smaller number of images. When manually annotated and automatically extracted parameters are compared,a dice coefficient of 92.091% for segmentation is obtained,as well as an average root mean square error (RMSE) of 0.3052cm for parameter extraction in apical view images and an average RMSE of 1.3548cm for parameter extraction in M-mode view images. Based on extracted left ventricle length,radiuses,wall thicknesses,and user-supplied divisions,3D parametric model of LV is reconstructed,which can be used to replicate the entire cardiac cycle using Finite element method (FEM). Described methodology is integrated into SILICOFCM platform [2]. Fully automated cardiomyopathy identification,3D reconstruction,and cardiac cycle modeling of the left ventricle utilizing ultrasound pictures can assist doctors in making more timely judgments and establishing more accurate therapies. The approach is now available on a user-friendly platform in order to help clinicians make faster decisions and establish reliable treatments.