Key result
Applying multiple image augmentation techniques to ECG images reduced deep learning classification accuracy from 0.818 to 0.764 and F1 score from 0.776 to 0.768 compared to no augmentation.
Why the study?
Deep learning requires large datasets, and the impact of using image augmentation on ECG images for detecting COVID-19 and heart diseases needed to be examined.
Does image augmentation improve the detection accuracy of deep learning models for ECG image classification?
Does image augmentation improve the detection accuracy of deep learning models for ECG image classification?
Applying multiple image augmentation techniques can adversely impact the accuracy of deep learning models for ECG classification.
Caution against multiple augmentations in ECG DL models; leaves open optimal preprocessing for cardiac imaging AI.
The advent of artificial intelligence has led to a better investigation of many complex research problems from a variety of domains. Recently deep learning approaches have emerged as cutting edge AI technologies and has been proved very effective in medical research. A large number of studies have recently used deep learning in the field of medical imaging for the detection and identification of various diseases including COVID19. In this work, we have investigated an ECG based approach for detection of COVID-19 and heart diseases using deep learning. Deep learning requires a lot of data and image augmentation is a way to enhance the size of data. In this study, we specifically examined the impact of augmenting ECG images for disease detection. Our study indicates that augmentation improves the detection accuracy to a certain extent and can adversely impact beyond that. Without augmentation, we achieved accuracy and F1 score of 0.818 and 0.776 which is reduced to 0.764 and 0.768 respectively, when multiple augmentation techniques are applied.
No takes yet. Share an insight, caveat, or question.
Anwar et al. (2021) studied COVID-19 and Cardiovascular Diseases (n=1,937). Image augmentation vs. No augmentation was evaluated on Classification accuracy. Applying multiple image augmentation techniques to ECG images reduced deep learning classification accuracy from 0.818 to 0.764 and F1 score from 0.776 to 0.768 compared to no augmentation.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: