Year: 2024
Type: conference-paper
Abstract: As part of the Digitization and Classification of ECG Images: The George B. Moody PhysioNet Challenge 2024, we developed a deep learning-based approach to classify electrocardiogram (ECG) images.Our methodology utilized the ConvNext architecture pre-trained on ECG images from CODE15 dataset, along with our private imagebased ECG datasets, InCor-AMB and InCor-EMG.We fine-tuned the model on the PTB-XL dataset to classify 11 specific abnormalities.The training strategy included cross-entropy loss minimization, class-weight adjustments, and the use of the AdamW optimizer over 30 epochs, with early stopping to mitigate overfitting.A 5-fold crossvalidation was employed, and the final model was an ensemble of the best-performing models from each fold.Our model achieved an F1-score of 0.698 ± 0.022 on internal 5-fold cross-validation.On the challenge's test set, our team, AIMED, achieved an F1-score of 0.817, securing 1 st place in the classification task of the PhysioNet Challenge 2024.This study highlights significant advancements in the integration of digital technologies in clinical settings, with the potential to improve ECG diagnostic accuracy worldwide.
Source: Computing in cardiology
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4405435409