Automatic Diagnosis of Cardiac Disease from Twelve-Lead and Reduced-Lead ECGs Using Multilabel Classification

ECG is an essential tool for the clinical diagnosis of cardiac electrical abnormalities. As part of the PhysioNet/Computing in Cardiology Challenge 2021, eight and two folds from the 10-folds iterative splitting of public training data set were used as in-house training and internal validation sets. We used extracted features from RandOm Convolutional KErnel Transforms (ROCKETs) with a multilabel classification using XGBoost to predict cardiac abnormalities. Our team, LINC, developed an approach with minimal pre-processing (e.g., resampling data to 500Hz) and with no QRS detection or deep neural network design, which led to promising performance on the internal validation set. We didn't receive the official scores for the validation and test sets, because our entry failed during training in the official phase as we submitted an incomplete entry. Our classifiers received scores of 0.504, 0.466, 0.459, 0.458, and 0.438 for the 12-lead, 6-lead, 4-lead, 3-lead, and 2-lead versions on the internal validation set with the challenge evaluation metric (10 seconds ECG).