Fig. 5: Model confusion patterns and demographic risk distribution.
From: Diagnosis of cardiac conditions from 12-lead electrocardiogram through natural language supervision
a False positive confusion matrix showing the most frequent misclassifications for each cardiac condition. Rows represent false positive conditions and columns represent true positive conditions. Circles indicate confusion frequency ranking (orange = most frequent, purple = second most frequent, blue = third most frequent); triangles represent co-occurring false positive conditions. b Model confusion risk ranking across sex-age demographic groups for each target condition, showing the top 10 highest-risk subgroups. Numbers represent risk rank (1 = highest risk, 10 = lowest risk among top 10); colors indicate risk level (dark red = ranks 1–3, orange = ranks 4–7, yellow = ranks 8–10). Cardiac condition abbreviations as in Fig. 3.
