CPSC2021

Paroxysmal Atrial Fibrillation Events Detection from Dynamic ECG Recordings: The 4th China Physiological Signal Challenge 2021

Offline generated training data

CPSC2021-sliced on Kaggle.

Graphical Abstract of the Solution

Main files

Click to view!

• Files required by the submission system
  • entry_2021.py: entry file of the challenge
  • score_2021.py: scoring function of the challenge
• Files of the solution:
  • cfg.py: configurations for the preprocessing, augmentation, model, trainer, etc.
  • data_reader.py: contains a class which makes reading data, annotations, etc. more easy
  • dataset.py: torch Dataset which generates tensors for training the models
  • model.py: models, adjustification of the architecture is highly flexible
  • trainer.py: contains a trainer class
• Files for analyzing and visualizing experiments:
  • gather_results.py
• Test files:
  • test_entry.py
• Notebooks (the filenames imply the usages):
  • inspect_data.ipynb
  • train_models.ipynb
  • aggregate_val_res.ipynb

Remember to change BaseCfg.db_dir to your local data directory.

Results (Rankings)

Results on the hidden test set

Click to view!

Name	Affiliation	Team Members	Test_I	Test_II	Average
usstmed	University of Shanghai for Science and Technology	Wenjie Cai, Fanli Liu, Bolin Xu, Xuan Wang, Yufeng Ji	2.0629	4.3921	3.2275
CeZIS	1.VSL Software, a.s., Košice, Slovakia; 2.Pavol Jozef Šafárik University in Košice, Slovakia	Peter Bugata1, Peter Bugata Jr.1, David Gajdos1, David Hudak1,Vladimira Kmecova1, Monika Stankova1, Lubomir Antoni2, Erik Bruoth2, Simon Horvat2, Richard Stana2, Alexander Szabari2, Gabriela Vozarikova2	2.1128	3.5716	2.8422
UNIWA	University of West Attica, Greece	Lampros Kokkalas, Nicolas A. Tatlas, Stelios M. Potirakis	2.0144	3.3833	2.6989
lingshui_BME	1.Dalian University of Technology; 2.RWTH Aachen University	Yating Hu1, Tengfei Feng2, Hong Tang1	1.8754	3.5116	2.6935
Taozi	Tianjin Medical University, Beihang University	Jingsu Kang, Hao Wen	1.9972	3.0907	2.5440
Metformin-121	National Taiwan University; Academia Sinica; Taiwan Artificial Intelligence Academy Foundation	Tsai-Min Chen, Yi-Dar Tang, Huan-Hsin Tseng, Wei Luok Ngu, Le-Yin Hsu, Miao-Chen Chiang, Yu-Te Ku, Ming-Yi Hong, Yu Tsao	1.6277	2.6649	2.1463
MVTECH	Shanghai Medical Vision Technology Co. Ltd.	Yang Hou, Jinlei Li	1.7966	2.4346	2.1156
FUDU_Car	Fudan University	Sen Liu, Ya‘nan Wang, Haijun Jia	1.9147	1.9473	1.9310
WHS	1.Central South University 2.China University of Geosciences	Lebing Pan, Jiechen Tang	1.8585	1.9236	1.8911
Muhammad Uzair Zahid	1. Tampere University, Finland; 2. Qatar University, Qatar	Muhammad Uzair Zahid1, Mustafa Serkan Kiranyaz2, Moncef Gabbouj1	1.8192	1.9469	1.8831
DaBin	Beijing University of Technology	Fengya Liu, Rui Yu, Shuicai Wu, Guangyu Bin,Zhuhuang Zhou ,Qian Wang	1.8352	0.9728	1.4040
Lastone	King's College London, UK	Xinqi Bao, Fenghe Hu	0.9616	1.6430	1.3023
BSU工程小分队	Beijing Sport University	Kuan Tao, Lixin Sun	1.2384	1.1710	1.2047
CUTCM	City University of Hong Kong	Marshall	0.6706	1.3854	1.0280
AZ-unet	AstraZeneca	Hannes Whittingham, Long Luu	1.1168	0.7788	0.9478
AIBI_LAB	Ludong University	Shuhong Wei, Yipeng Wang, Yu Ji, Yinhao Sun	0.7837	1.1006	0.9422
CPSC_eie	Zhejiang University of Technology, China	Xinyuan Ying, Qing Pan, Ziyou Zhang	0.8555	0.9629	0.9092
Baseline	/	/	0.6819	0.6485	0.6652

See the official website for more details.

Results on the validation set

Raw results are gathered into one zip file, the val_res.zip in the results folder

Network(s)	Merge Rule	Score on Partial Hidden Test Set	Score on Validation Set
LSTM	NA	1.9392	2.0621
SeqTag	NA	1.9781	2.1577
U-Net	NA	1.3699	NA
LSTM + U-Net	Union	1.7829	NA
LSTM + SeqTag	Intersection	1.9287	NA
LSTM + SeqTag	Union	1.9766	2.1682
LSTM + SeqTag	New Union	1.9972	2.2179

Confusion matrices of the LSTM model and the SeqTag model

              

More detailed analysis using pandas

Citation

See the file CITATIONS.bib. Also at sciencedirect.

References

See the references listed on the sciencedirect webpage.