Performance of Vision-Enabled Large Language Models in Image-Based Electrocardiogram Interpretation: Exploratory Evaluation
Nibras Soubh, Eva Rasenack, Helge Haarmann, Felix Wiedmann, Markus Zabel, Constanze Schmidt, Rayan Suliman, Leonard Bergau · 2026
Read Performance of Vision-Enabled Large Language Models in Image-Based Electrocardiogram Interpretation: Exploratory Evaluation by Nibras Soubh, Eva Rasenack, Helge Haarmann, Felix Wiedmann, Markus Zabel, Constanze Schmidt, Rayan Suliman, Leonard Bergau (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13234008.txt.
Description
Nibras Soubh and colleagues published a 2026 evaluation of vision-capable language models reading electrocardiogram images. Six leading generalist models, plus two ECG-specialized ones, were tested on seventy real patient ECG tracings stripped of identifiers. Each model had to answer nine standard questions about rhythm, conduction blocks, intervals, extra beats, and signs of ischemia, judged against expert consensus.
Generalist accuracy was mediocre, with balanced scores between about fifty and sixty-two percent. The best model was also by far the slowest. Sensitivity for atrial fibrillation was dismal, at best catching one of nine cases, and detection of conduction delays and interval prolongation was likewise poor. The authors warn that today's general-purpose vision models are nowhere near safe for independent ECG interpretation.
Catalog record
| CREATOR | Nibras Soubh, Eva Rasenack, Helge Haarmann, Felix Wiedmann, Markus Zabel, Constanze Schmidt, Rayan Suliman, Leonard Bergau |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 67 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |