Table 2 Classification performance on PTB-XL benchmarking tasks²⁶ (macro AUC on the PTB-XL test set) achieved using different feature sets using different PTB-XL label (sub)sets as targets (all: all 71 statements, diag: 44 diagnostic statements, sub-diag: 23 aggregated, sub-diagnostic statements, super-diag: 5 aggregated, super-diagnostic statements, form: 19 form-related statements, rhythm: 12 rhythm-related statements).

Model/Features	all	diag	sub-diag.	super-diag.	form	rhythm
CNN/raw data²⁶	0.925	0.937	0.929	0.928	0.896	0.957
RF/Uni-G(full)	0.875	0.907	0.886	0.921	0.803	0.945
RF/12SL(full)	0.856	0.906	0.878	0.924	0.794	0.870
RF/ECGDeli(full)	0.864	0.891	0.883	0.899	0.776	0.964
RF/Uni-G(Uni-G ∩ 12SL)	0.855	0.890	0.889	0.923	0.773	0.881
RF/12SL(Uni-G ∩ 12SL)	0.866	0.892	0.881	0.922	0.796	0.860
RF/Uni-G(Uni-G ∩ ECGDeli)	0.863	0.902	0.888	0.916	0.769	0.892
RF/ECGDeli(Uni-G ∩ ECGDeli)	0.855	0.898	0.863	0.902	0.753	0.906
RF/12SL(12SL ∩ ECGDeli)	0.857	0.894	0.877	0.919	0.781	0.872
RF/ECGDeli(12SL ∩ ECGDeli)	0.855	0.884	0.889	0.903	0.764	0.902

Best-performing feature-based approaches in each category are marked in bold face. Overall best-performing approaches are underlined.

Quick links

Search