Machine learning to predict no reflow and in-hospital mortality in patients with ST-segment elevation myocardial infarction that underwent primary percutaneous coronary intervention

Figures

• 

  Fig. 1. Schematic workflow for AI. The AI method to predict NR and in-hospital mortality included feature selection, training, and testing split, which combined four different model algorithms (e.g., RAN, NNET, SVM, and CTREE). Random sampling was conducted 10 times, and the results were averaged. AI, artificial intelligence; CTREE, Decision Tree; NNET, Neural Network Algorithm; RAN, random forest; SVM, Support Vector Machine

• 

  Fig. 2. Correlation analysis of the variables in the dataset to predict NR, as shown in the heat map, the dataset was not composed of many correlated variables, which indicated the model was simpler and more stable. ALB, albumin; Cr, creatinine; CK-MB, creatine kinase-MB; cTnI, cardiac troponin I; DM, diabetes mellitus; D-to-B, door-to-balloon; EF, ejection fraction; Fib, Fibrinogen; Glu, glucose; Hb, hemoglobin; Hbp, high blood pressure; HDL-C, high-density lipoprotein cholesterol; hs-CRP, high-sensitivity C-reactive protein; LDL-C, low-density lipoprotein cholesterol; PCI, percutaneous coronary intervention; SO-to-FMC, symptom-onset to-first medical contact; TC, total cholesterol; TG, triglyceride; TIMI, thrombolysis and thrombin inhibition in myocardial infarction; UA, uric acid; WBC, white blood cells

• 

  Fig. 3. Correlation analysis of the variables in the dataset for the prediction of in-hospital mortality, the dataset was not composed of many correlated variables, which indicated the model was simpler and more stable. ALB, albumin; Cr, creatinine; CK-MB, creatine kinase-MB; cTnI, cardiac troponin I; DM, diabetes mellitus; D-to-B, door-to-balloon; EF, ejection fraction; Fib, Fibrinogen; Glu, glucose; Hb, hemoglobin; Hbp, high blood pressure; HDL-C, high-density lipoprotein cholesterol; hs-CRP, high-sensitivity C-reactive protein; LDL-C, low-density lipoprotein cholesterol; PCI, percutaneous coronary intervention; SO-to-FMC, symptom-onset to-first medical contact; TC, total cholesterol; TG, triglyceride; TIMI, thrombolysis and thrombin inhibition in myocardial infarction; UA, uric acid; WBC, white blood cells

• 

  Fig. 4. ROC curves to predict NR and in-hospital mortality: A a RAN model presented a higher AUC for NR; and B in-hospital mortality prediction than all other models (e.g., NNET, SVM, and CTREE). AUC, area under the curve; CTREE, Decision Tree; NNET, Neural Network Algorithm; RAN, random forest; SVM, Support Vector Machine

• 

  Fig. 5. Ranking of feature importance to predict NR: Contribution of clinical, laboratory, and angiographic variables to the optimal AI-based prediction model (RAN). ALB, albumin; Cr, creatinine; CK-MB, creatine kinase-MB; cTnI, cardiac troponin I; DM, diabetes mellitus; D-to-B, door-to-balloon; EF, ejection fraction; FIB, Fibrinogen; Glu, glucose; Hb, hemoglobin; HBP, high blood pressure; HDL-C, high-density lipoprotein cholesterol; hs-CRP, high-sensitivity C-reactive protein; LDL-C, low-density lipoprotein cholesterol; MA, Malignant arrhythmia; PCI, percutaneous coronary intervention; RAN, random forest; SO-to-FMC, symptom-onset to-first medical contact; TC, total cholesterol; TG, triglyceride; TIMI, thrombolysis and thrombin inhibition in myocardial infarction; UA, uric acid; WBC, white blood cells

• 

  Fig. 6. Ranking of feature importance to predict in-hospital mortality: Contribution of clinical, laboratory, and angiographic variables to the optimal AI-based prediction model (RAN). ALB, albumin; Cr, creatinine; CK-MB, creatine kinase-MB; cTnI, cardiac troponin I; DM, diabetes mellitus; D-to-B, door-to-balloon; EF, ejection fraction; FIB, Fibrinogen; Glu, glucose; Hb, hemoglobin; HBP, high blood pressure; HDL-C, high-density lipoprotein cholesterol; hs-CRP, high-sensitivity C-reactive protein; LDL-C, low-density lipoprotein cholesterol; MA, Malignant arrhythmia; PCI, percutaneous coronary intervention; RAN, random forest; SO-to-FMC, symptom-onset to-first medical contact; TC, total cholesterol; TG, triglyceride; TIMI, thrombolysis and thrombin inhibition in myocardial infarction; UA, uric acid; WBC, white blood cells

• 

  Fig. 7. Model validation between different centers, one center (The Second People's Hospital of Nanning) as a training set and another center (The First People's Hospital of Nanning) as a testing set, ROC curves to predict NR and in-hospital mortality: A a RAN model presented a higher AUC for NR; and B in-hospital mortality prediction than all other models (NNET, SVM, and CTREE). AUC, area under the curve; CTREE, Decision Tree; NNET, Neural Network Algorithm; RAN, random forest; SVM, Support Vector Machine

• 

  Fig. 8. Statistical analysis was performed by controlling demographic factors. Machine learning procedure was performed after removal of age and gender. ROC curves to predict NR and in-hospital mortality: A a RAN model presented a higher AUC for NR and; B in-hospital mortality prediction than all other models (NNET, SVM, and CTREE). Texture feature selection using the least absolute shrinkage and selection operator (Lasso) binary logistic regression model. The tuning parameter (λ) selection in the Lasso model was based on minimum criteria. C Lasso coefficient profiles of the 36 features for NR; D Lasso coefficient profiles of the 37 features for in-hospital mortality; coefficient profile plot was produced against the log (λ) sequence; E AUC curve was plotted versus log (λ) for NR; and F in-hospital mortality. Dotted vertical lines were drawn at the optimal values using the minimum criteria and one standard error of the minimum criteria. AUC, area under the curve; CTREE, Decision Tree; NNET, Neural Network Algorithm; RAN, random forest; SVM, Support Vector Machine