Explainable Deep Neural Network for Identifying Cardiac Abnormalities Using Class Activation Map
Journal
Computing in Cardiology
Journal Volume
2020-September
ISBN
9.78173E+12
Date Issued
2020
Author(s)
Abstract
In this study, our team 'NTU-Accesslab' present a deep convolutional neural network (CNN) approach, called CNN-GAP, for classifying 12-lead ECGs with multilabel cardiac abnormalities. Additionally, Class Activation Mapping (CAM) is employed for further understanding the decision-making process of this black-box model, making the model more explainable. The CNN-GAP model consists of 12 layer Conv Blocks along with Batch Normalization layer, Global Average Pooling and Fully Connected layer with sigmoid activation. To deal with the data imbalance problem, we oversample the minor datas. In the training stage, we applied Macro observed score loss (Macro-Obs) instead of the conventional Weighted Cross entropy loss (WCE), and we have shown that this results in higher challenge scores. Additionally, we augmented datas by randomly scaling datas to get better scores and prevent model overfitting. Our method achieved a challenge score of 0.58 on the validation set, but was unable to score and rank on the test set, due to a failure of the algorithm on the fully hidden dataset. © 2020 Creative Commons; the authors hold their copyright.
SDGs
Other Subjects
Cardiology; Chemical activation; Convolutional neural networks; Decision making; Statistical tests; Activation mapping; Black-box model; Cross entropy; Data imbalance; Decision making process; Multi-label; Overfitting; Oversample; Deep neural networks
Type
conference paper