Evaluation procedure

Protocol

According to the recommendation of the American Society of Echocardiography and the European Association of Cardiovascular Imaging, manual contours were drawn on apical four-chamber and two-chamber views at End diastole (ED) and End Systole (ES). Since some patients involved in the CAMUS dataset had non-exploitable ECG due to clinical conditions, in accordance with the recommendations, ED (respectively ES) was defined as the frame in the cardiac cycle in which the left ventricle dimension is the largest (respectively the smallest). While only the extraction of the left ventricle endocardium (LV Endo) contour is necessary to estimate left ventricle ejection fraction values, we also asked the cardiologists to manually outline the left ventricle epicardium (LV Epi) and the left atrium (LA) for all patients. This was done to study the influence of contextualization (segmentation of several structures at once) on the performance of the left ventricle endocardium segmentation. The following protocol was set up.

• LV Endo : Convention was used for the LV wall, mitral valve plane, trabeculations, papillary muscles and apex. Basic points were to i) include trabeculae and papillary muscles in the LV cavity; ii) keep tissue consistency between ED and ES instants; iii) terminate the contours in the mitral valve plane on the ventricular side of the bright ridge, at the points where the valve leaflets are hinging; iv) partially exclude left ventricular outflow tract from the cavity by drawing from septal mitral valve hinge point to the septal wall to create a smooth shape.


• LV Epi : There is no recommendation for delineating the epicardium. We thus outlined the epicardium as the interface between the pericardium and the myocardium for the anterior, anterolateral and inferior segments and the frontier between the right ventricle cavity and the septum for the inferoseptal segments.


• LA: There are recommendations for LA segmentation to assess the full LA area from dedicated LA recordings. However, since we have used acquisitions focusing on the LV, part of the dataset does not cover the full LA surface and is thus not suited to perform such measurement. Having this in mind, we used the following contouring protocol: i) start the LA contour from the extremities of the LV Endo contour, at the points where the valve leaflets are hinging; ii) have the contour pass by the LA inner border.