MTFNet: A Morphological and Temporal Features Network for multiple leads ECG Classification

Lebing Pan, Weibai Pan, Mengxue Li, Yuxia Guan, Ying An
Central South University


Cardiac abnormalities are the main cause of death and ECG is a key diagnostic tool to assess the cardiac condition of a patient. The goal of the 2021 PhysioNet/CinC Challenge was to develop algorithms to diagnose multiple cardiac abnormalities from reduced-lead ECG. In this work, We describe a new model that extracts morphological and temporal features to classify each ECG sequence into 26 cardiac abnormality classes. We use multiple double conv blocks with multiple layers of small convolution kernels to extract morphological features, each layer has similar hyperparameters. Then the Bidirectional LSTM is used to obtain the temporal characteristics. Our entry to the 2021 PhysioNet/CinC Challenge, using the official generalized weighted accuracy metric for evaluation, Our classifiers received scores of 0.61, 0.59, 0.60, 0.60, and 0.59 (ranked 14th, 12th, 12th, 13th, and 11th out of 60 teams) for the 12-lead, 6-lead, 4-lead, 3-lead, and 2-lead versions of the hidden validation set.