ECG-image model identifies concurrent low ejection fraction in archived records
Source check: Source checked. Not independent scientific confirmation.
Finding: A DenseNet121 model learned to identify echocardiography-referenced ejection fraction below50% from ECG images. It reported ROC areas of0.92 on an internal ECG test set and0.88 at another healthcare centre. The study tests concurrent diagnostic-label discrimination, without establishing early warning or clinical benefit.
Technology: DenseNet121 convolutional neural network
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- A trained convolutional network maps cropped ECG trace images to a concurrent echocardiographic low-ejection-fraction label.
- Test and result
- Development used119,281 same-day ECG/echo pairs from104,697 people. ECGs were randomly split into107,147 training,2,020 validation and10,114 test records. Another centre supplied roughly24,300 pairs; the internal-test Youden threshold0.18 was carried to that cohort.
- Limitation / still unknown
- Random ECG-level splitting does not establish separation of repeated patients. The external denominator is24,323 in text versus24,319 in the table. The internal threshold was selected on test outcomes; the external data are archived, not a prospectively frozen workflow. Standardized device images and same-day selected imaging pairs limit generalization; no screening benefit or diagnosis of future dysfunction is established.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12223524/ · https://doi.org/10.1016/j.ihj.2025.03.009
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 21 Nov 2026. Review scope: Original primary clinical/computational Methods, reference definitions, model evaluation and Results independently read; no operational laboratory sections.
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.