Table 2.
Summary of the main cited studies providing an overview of the role of machine learning in the field of CRT.
| Authors/Journal – year/Ref | Population | Endpoints | Methods | Parameters | Results |
|---|---|---|---|---|---|
| Donal E et al. JACC Cardiovasc Imaging 2019. (75) | 154 pts LVEF < 35%, QRS width > 120 msec |
LVESV reduction >15% at 6 months FU | Random Forest and Monte Carlo cross-validation | 60 features from semi-automatic strain trace analysis, QRS width, visual Lv dyssynchrony | Extraction of 6 main features with an AUC 0.804 (0.77–0.94) for the prediction of LV reverse remodelling |
| Gallard A. et al. Int J Cardiovasc Imaging 2021. (76) | 221 pts LVEF < 35%, QRS ≥120 msec |
LVESV reduction >15% at 6 months FU | Random Forest | Features from echocardiography, strain traces analysis, QRS width, LV dyssynchrony | The combination of SF, E, E/A, E/e′, QRS width, LVESV and 8 features obtained from strain curves predicted LV reverse remodelling with an AUC of 0.81 ± 0.05 |
| Gallard A et al. PloS One 2021. (77) | 161 pts LVEF < 35%, QRS width > 120 msec |
LVESV reduction >15% at 6 months FU | Random Forest | 158 features from semi-automatic strain trace analysis, QRS width, LVEF | 20 main features from strain analysis are the best predictors of LV reverse remodelling. 50% of these features are derived from the 4-chamber strain. |
| Cikes M et al. Eur J Heart Fail 2019. (78) | 1,106 LVEF ≤ 30%, QRS width ≥130 ms |
Overall death or HF hospitalisation | Multiple Kernel Learning and K-means clustering | Four phenogroups | Two phenogroups with better prognosis (HR = 0.35, 95% CI: 0.19–0.64, p < 0.0001 and HR = 0.36, 95% CI: 0.19–0.68, p = 0.001) |
| Galli E et al. J Am Soc Echocardiogr 2021. (79) | 193 pts LVEF 29 ± 8%, QRS width 167 ± 21, 33% ischemic |
Primary: heart transplantation, LV-assisted device implantation overall death | k-medoid, Boruta algorithm, random forest | 28 clinical, biological and echo-derived variables; two phenogroups | Prognostic value of the main 8 variables: AUC 0.84 (0.75–0.93); Two phenogroups with different prognosis: HR = 4.70, 95% CI: 2.1–10.0, p < 0.0001; log-rank p < 0.0001) |
| Bivona DJ et al. Heart Rhythm 2022. (80) | 200 pts LVEF 24.0 (17.7–30.5)%, QRS 58 (142–175) msec, 43.5% ischemic |
1) Reduction of LVESV at 6 months FU 2) Survival |
k-means method and Gaussian mixture model | 39 features derived from CMR and biology | 3 clusters able to predict CRT response and prognosis |
AUC, area under the curve; CI, confidence interval; CMR, cardiac magnetic resonance; FU, follow-up; HF, heart failure; HR, hazard ratio; LV, left ventricle; LVEF, left ventriculare ejection fraction; LVESV, left ventricular end-systolic volume.