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_2021.py): entry file of the challenge - [`score_2021.py`](score_2021.py): scoring function of the challenge * Files of the solution: - [`cfg.py`](cfg.py): configurations for the preprocessing, augmentation, model, trainer, etc. - [`data_reader.py`](data_reader.py): contains a class which makes reading data, annotations, etc. more easy - [`dataset.py`](dataset.py): torch Dataset which generates tensors for training the models - [`model.py`](model.py): models, adjustification of the architecture is highly flexible - [`trainer.py`](trainer.py): contains a trainer class * Files for analyzing and visualizing experiments: - [`gather_results.py`](gather_results.py) * Test files: - [`test_entry.py`](test_entry.py) * Notebooks (the filenames imply the usages): - [`inspect_data.ipynb`](inspect_data.ipynb) - [`train_models.ipynb`](train_models.ipynb) - [`aggregate_val_res.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.