For example,Бобцов

SOFTWARE PACKAGE FOR 17-SEGMENT LOCALIZATION OF MYOCARDIAL FIBROSIS OF THE LEFT VENTRICLE OF THE HEART

Annotation

Localized quantitative assessment of myocardial fibrosis is used to predict outcomes of cardiovascular diseases, select therapeutic and surgical strategies. In widespread practice, localized quantitative assessment of myocardial fibrosis is complicated by the need for manual or semiautomatic image preprocessing. A software package based on the use of deep learning technologies for analyzing MR images of the heart with late gadolinium enhancement and quantifying myocardial fibrosis is proposed. The results of statistical analysis of the prepared dataset indicate the presence of stable correlations of localized relative fibrosis volume with clinical data on ejection fractions and the degree of heart failure in patients. A pronounced negative correlation can be noted between the degree of heart failure and the left ventricular ejection fraction (p <0.001). There is also a small correlation between the LV ejection fraction and the relative volume of fibrosis for two basins - RCA and LCx (p < 0.01). The coefficient of structural similarity with the reference marking of the myocardium is 0.87 / 0.87 / 0.88 according to the DSC / Precision / Recall metrics for the automatic Cascade-U-Net-based solution. The accuracy of classification of the left ventricle level on the slice by the trained U-NetLoc model is 83% / 86% / 95% in the classification of basal / medial / apical myocardial slices. The quality of myocardial localization by the trained U-NetBull model ensures the Sorensen–Dice similarity coefficient 0,827/0,778/0,734 in individual segments of the basal / medial / apical level of the left ventricle level. The developed software package for localized quantitative assessment of myocardial fibrosis can be used as an auxiliary tool by cardiologists and radiologists to increase the diagnostic accuracy.

Keywords

Articles in current issue