Mixup Asymmetric Tri-training for Heartbeat Classification Under Domain Shift

In this paper, we present MIAT, a novel UDA-based method for heartbeat classification aims to reduce the domain shift issue by integrating asymmetric tri-training and three kinds of mixup regularizations.

Main requirements

• Torch == 1.0.0
• Python == 3.5
• WFDB == 1.2.2

Task

Classify ECG heartbeats into 5 classes: N, S, V, F, Q

Dataset

Download MIT-BIH Arrhythmia Database (MITDB)
Download MIT-BIH Supraventricular Arrhythmia Database (SVDB)

Usage

(1) Cut the continuous ECG signals into heartbeat segments:
python data_process_for_MIAT.py

(2) train and evaluation task DS1->DS2:
python MIAT_train_eval.py --run_id=0 --gpu=0 --epochs=150 --lr=0.001 --weight=0.005 --n=5 --lambda=1 --alpha=2 --prevat=1 --mix=1 --vat=1 --s=DS1 --t=DS2

(3) train and evaluation task MITDB->SVDB:
python MIAT_train_eval.py --run_id=5 --gpu=2 --epochs=150 --lr=0.001 --weight=0.005 --n=1 --lambda=5 --alpha=1 --prevat=1 --mix=1 --vat=1 --s=mitdb --t=svdb