The Impact of Domain Shift on Left and Right Ventricle Segmentation in Short Axis Cardiac MR Images

Domain shift refers to the difference in data distribution of two datasets, normally between training set and test for machine learning algorithms. is a serious problem generalization models it well-established that domain sets may cause drastic drop model’s performance. In medical imaging, there can be many sources such as different scanners or scan protocols, pathologies patient population, anatomical differences population (e.g. men vs women) etc. Therefore, order train have good performance, important aware problem, its potential causes devise ways address it. this paper, we study effect on left right ventricle blood pool segmentation short axis cardiac MR images. Our dataset contains images from 4 3 pathology groups. The performed with nnUNet. results show scanner greater performance compared changing group, impact segmentation. Increasing number subjects increased cross-scanner more than in-scanner at small sizes, but improvement decreased larger sizes. Training using multiple improved cross-domain