Table 3 Additional prognostic contribution of the proposed enhanced LV function characterization.

Model	Linear selection	AUC_k	AUC_RS	Cox selection	C_k	C_RS
LVEF	LVEF	0.683 (0.681—0.685)	0.687	LVEF	0.669 (0.668—0.670)	0.669
CMR	ESV, EDV	0.700 (0.698—0.702)	0.708	ESV, EDV	0.688 (0.687—0.690)	0.693
CMR + Vt	ESV, EDV, RR, t_diastolic	0.708 (0.706—0.710)	0.721	ESV, EDV	0.688 (0.687—0.689)	0.693
CMR + Vt_AI	ESV, EDV, Vt_AI2, Vt_AI3, Vt_AI5	0.726 (0.723—0.730)	0.742	ESV, EDV, Vt_AI3, Vt_AI5	0.704 (0.702—0.707)	0.716
CMR + Strain	ESV, EDV, GLS	0.723 (0.721—0.725)	0.733	ESV, EDV, GLS	0.710 (0.708—0.711)	0.717
CMR + LV_3D	ESV, EDV, C5, C16	0.738 (0.736—0.740)	0.750	ESV, EDV, C5, C16	0.728 (0.727—0.730)	0.736
ALL	ESV, EDV, Age, Killip	0.729 (0.727—0.733)	0.745	ESV, EDV, Age, Killip	0.714 (0.712—0.717)	0.726
ALL + Vt	ESV, Age, Killip, BSA, RR, t_diastolic	0.737 (0.734—0.740)	0.759	ESV, Age, Killip, RR, v_{diastolic_avg}	0.717 (0.714—0.721)	0.732
ALL + Vt_AI	ESV, EDV, Age, Killip, Vt_AI3, Vt_AI5	0.746 (0.743—0.749)	0.769	ESV, EDV, Age, Killip, Vt_AI3, Vt_AI5	0.730 (0.727—0.732)	0.747
ALL + Strain	ESV, Age, Killip, BSA, Weight, Vessels, GLS	0.752 (0.749—0.754)	0.776	ESV, Age, Killip, BSA, GLS	0.732 (0.730—0.736)	0.748
ALL + LV_3D	ESV, EDV, Age, Killip, C5, C16	0.747 (0.745—0.749)	0.766	ESV, EDV, Age, C5, C16	0.741 (0.739—0.743)	0.753
All + Vt_AI + Strain + LV_3D	ESV, EDV, Age, C16, GLS, Vt_AI3, Vt_AI5	0.774 (0.771—0.777)	0.796	ESV, EDV, Age, C16, GLS, Vt_AI5	0.760 (0.759—0.763)	0.773

Backward stepwise LDA and Cox results of additional prognostic metrics for MACE risk assessment. Considered features include conventional (Vt) and AI-derived (Vt_AI) volume transient metrics, CMR-FT strains (strains), and LV 3D detail patterns (LV_3D) descriptors, each on top of considering only CMR biomarkers (CMR) or the cardiovascular risk factors and patient characteristics of the study (ALL). A final experiment combines the 3 proposed directions of enhancement. Performance of LVEF is likewise included. The resulting significant selection of variables is reported along with performance, expressed as AUC and C-index (C) re-substitution (rs) and tenfold cross-validated (k), computed for a 100 random data splits and presented as median (interquartile range). All differences in AUC_k and C_k between models were statistically significant (P < 0.001), except for the ‘CMR’ and ‘CMR + Vt’ resulting Cox models. BSA indicates body surface area (m²); Killip, Killip class on admission; RR, RR-interval (ms); t_diastolic, diastolic time (ms); v_{diastolic_avg}, average diastolic velocity (mL/s); Vessel, number of diseased vessels.
Best performing results are highlighted in bold.

Search