New Radiomic Markers of Pulmonary Vein Morphology Associated With Post-Ablation Recurrence of Atrial Fibrillation

Figures

• 

  FIGURE 1.. Summary of patient selection criteria for inclusion in this study.

• 

  FIGURE 2.. Overview of experiment methodology from data and radiomic feature extraction to selection, classification, and validation on an independent cohort.

• 

  FIGURE 3.. Summary of top five radiomic and top five clinical features each selected by method with 5-fold cross validation for all, radio-frequency, paroxysmal AF, and persistent AF subject (other features not selected are not represented). AP = Anterior Posterior; LAV = Left Atrial Volume; LI = Left Inferior; LS = Left Superior; RI = Right Inferior; RS = Right Superior; PV = Pulmonary Vein; Hx = History; Ht = Height; LVEF = Left Ventricular Ejection Fraction; AF = Atrial Fibrillation.

• 

  FIGURE 4.. Classifier performance with cross-validation for post-PVI recurrence of AF within 1 year for D1 all subjects (n=150) with 16 combinations of feature selection and classifier models. Models were trained on radiomic features ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\text{M}_{\mathrm {R}}$ \end{document}), clinical features ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\text{M}_{\mathrm {C}}$ \end{document}), and the combination of radiomic and clinical features ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\text{M}_{\mathrm {RC}}$ \end{document}).

• 

  FIGURE 5.. Left: Summary of 1-year post-PVI AF recurrence, Wilcoxon-selected, radiomic/clinical ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\text{M}_{\mathrm {RC}}$ \end{document}) feature effect size by sub-cohort represented as log odds ratio in D1 subjects (n=150). Right: graphical representation of feature effects ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\text{M}_{\mathrm {R}}$ \end{document}) for all subjects in D1.

• 

  FIGURE 6.. Comparison of LDA model performances using Wilcoxon-selected features by feature type and cohort subset on D2 reveals a trend for improved model performance combining radiomic and clinical features ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\text{M}_{\mathrm {RC}}$ \end{document}), with best performance in subjects undergoing radio-frequency ablation. Radiomic- based models ( \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\text{M}_{\mathrm {R}}$ \end{document}) were unable to reliably predict recurrence in patients with persistent AF.