Core Concepts
ECG anomaly detection using multimodal time and spectrogram restoration network.
Abstract
ECG is crucial for heart health assessment.
Proposed TSRNet leverages anomaly detection for identifying unhealthy conditions.
Combines time series and time-frequency domain aspects for robust anomaly detection.
Introduces Peak-based Error for focusing on ECG peaks in anomaly detection.
TSRNet trained on normal ECG samples but tested on both normal and abnormal samples.
Outperforms other methods in ECG anomaly detection.
Compact model size and real-time inference capabilities.
Acknowledges contributions and funding sources.
Stats
"TSRNet achieves SOTA performance (AUC = 0.860)."
"Model size: 4.39M params."
"Inference speed: 33.3 fps."
Quotes
"TSRNet introduces an innovative perspective by emphasizing the substantial impact of both time-frequency and time-series domains on ECG anomaly detection."
"The experimental results demonstrate that TSRNet achieves SOTA performance (AUC = 0.860) while maintaining real-time inference capabilities (33.3 fps) and a compact model size (4.39M params)."