Deep neural networks learn by using human-selected electrocardiogram features and novel features
Key result
Deep neural networks match human ECG experts and generate novel features for superior performance.
Why the study?
It was unknown whether deep neural networks for ECG signal analysis can be explained using human-selected features, and whether AI models learn features similar to a human expert.
Population
ECG signals
Comparison
Deep neural networks vs human-selected features
Authors
Discussion
Loading...
Member takes
Implication
Supports AI-ECG tool development with human-like plus novel features; leaves open prospective validation before clinical use.
Structured PICO
P
Population
Electrocardiogram (ECG) signals
I
Intervention
Deep neural networks (NNs) for electrocardiogram (ECG) signal analysis
C
Comparator
Human-selected features / human experts
O
Outcome
Explainability of AI model and similarity of learned features to human experts
Deep neural networks for ECG analysis utilize both human-like features and novel features to achieve superior performance.
Cite This Study
Attia et al. (2021) studied ECG signal analysis. Deep neural networks (NNs) vs. Human experts was evaluated on Explainability and feature similarity to human experts. Deep neural networks for ECG signal analysis extract features similarly to human experts and generate additional novel features that help achieve superior performance (up to 0.86).