Machine Learning for Quality Assurance of Myocardial Strain Curves

Speckle tracking echocardiography (STE) is a well-established technique to quantify regional myocardial function. Reliability of STE-derived strain curves, however, depends strongly on the quality of the acquired B-mode images and can significantly be influenced by image artifacts. Artifactual images could lead to tracking errors and as a result, the measured deformation patterns might be similar to those obtained in pathology. It would thus be clinically very relevant to get feedback on the reliability (i.e. quality) of the extracted strain curves. As such, the aim of this study was to examine the utility of machine learning in the identification of artifactual strain curves. Our proposed learning framework was built upon a data imputation algorithm in order to facilitate the task of classifying (non-)artifactual curves. The obtained results confirmed the feasibility of automatic quality assurance of the STE-derived strain curves via machine learning.