Fig. 1

Sample features and learning algorithms. a 29 medical indices of one SLE patient were arranged in one column. The label was denoted in green when the SLE patient had a live birth and it was red when the patient had a fetal loss. However, these two classes had largely different quantities where only 38 patients labeled fetal loss in all the 338 patients. b Each sample was fed into the artificial neural network (ANN), where the learning rate was dependent on the sample label. c Every several samples were trained simultaneously in one batch in the ANN, where the learning rate positively depended on the number of fetal loss samples in the batch