Boosting ECG Classification Performance by Pre-training with Synthesized Data
Researchers have developed a method to generate synthetic ECG data using a knowledge-driven Gaussian-composition algorithm. This synthetic data was used to pre-train deep neural networks for classifying abnormal heart rhythms. The study found that pre-training with synthetic data improved classification performance for three out of four target abnormalities, with the most significant gains seen in atrial flutter detection. AI
IMPACT Synthetic data generation can overcome real-world data limitations, potentially accelerating AI adoption in specialized fields like medical diagnostics.