Abstract
Cardiac left ventricle (LV) quantification provides a tool for diagnosingcardiac diseases. Automatic calculation of all relevant LV indices from cardiacMR images is an intricate task due to large variations among patients anddeformation during the cardiac cycle. Typical methods are based on segmentationof the myocardium or direct regression from MR images. To consider cardiacmotion and deformation, recurrent neural networks and spatio-temporalconvolutional neural networks (CNNs) have been proposed. We study an approachcombining state-of-the-art models and emphasizing transfer learning to accountfor the small dataset provided for the LVQuan19 challenge. We compare 2Dspatial and 3D spatio-temporal CNNs for LV indices regression and cardiac phaseclassification. To incorporate segmentation information, we propose anarchitecture-independent segmentation-based regularization. To improve therobustness further, we employ a search scheme that identifies the optimalensemble from a set of architecture variants. Evaluating on the LVQuan19Challenge training dataset with 5-fold cross-validation, we achieve meanabsolute errors of 111 +- 76mm^2, 1.84 +- 0.9mm and 1.22 +- 0.6mm for area,dimension and regional wall thickness regression, respectively. The error ratefor cardiac phase classification is 6.7%.