Study tackles rare cardiac condition detection with imbalanced ECG data

Researchers have conducted a study on the detection of Wolff-Parkinson-White (WPW) syndrome, a rare cardiac condition, using a large dataset of electrocardiography (ECG) recordings. The study aimed to address the significant class imbalance, with only 142 WPW cases out of 66,951 recordings. Various detection methods were compared, including different signal representations and self-supervised pretraining, under a controlled protocol to prevent data leakage. The findings suggest that increased model diversity and capacity did not significantly improve detection performance, and even a feature-union model matched a simpler two-member vote. An error analysis revealed that missed cases often had narrower QRS complexes, and some false positives were mislabeled recordings. AI

IMPACT This research highlights challenges in applying AI to rare disease detection with imbalanced datasets, potentially informing future medical AI development.

RANK_REASON Research paper on a specific medical detection problem. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.LG →

• Chapman-Shaoxing-Ningbo
• PTB-XL
• Wolff–Parkinson–White syndrome

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →