PO-04-170 AUTOMATED CT-BASED PREDICTION OF VENTRICULAR ARRHYTHMIA

Abstract

Robust risk stratification of post-myocardial infarction (MI) ventricular tachycardia (VT) is desirable to overcome the limitations of left ventricular ejection fraction (LVEF), the currently used criterion for implanting cardiac defibrillator devices. We hypothesized, that a fully automated CT-based technology could predict VT in post-MI patients. A fully automated CT image processing pipeline was implemented including AI-based LV wall segmentation followed by joint analysis and quantification of wall thickness (WT), intramural fat, and myocardial calcification (Fig.1A). A cohort of 132 patients with prior MI (66±11 years, 37±10% LVEF) with implanted cardiac devices underwent CT scan, and were subsequently followed by remote monitoring over an average of 6.6 years (80 ± 31 months). We analyzed the relationship between CT markers (WT, fat, calcification) and VT events at follow-up. Analysis of the wall thickness layers showed the optimal WT-derived marker was the ratio between areas of WT < 3mm and WT< 5mm (no VT 0.57±0.45 vs. VT 0.89±0.55; p<0.01). The amount of myocardial calcification was also found higher in patients with VT (no VT 0.10±0.38ml vs. VT 0.47±1.57ml, p=0.039), while the amount of fat did not differ significantly between groups (no VT 1.14±1.38ml vs. VT 1.21±1.01ml, p=0.771). The optimal cut-off for the WT 3/5mm ratio was 0.3. On Kaplan Meier analysis, this WT ratio was a powerful predictor of VT-free survival (Fig.1B, log rank P<0.01). In contrast, a LVEF<35% was not found to significantly relate to VT-free survival (Fig.1C, P=0.77). Automated analysis of post-MI scar is doable from CT images, allowing for improved risk stratification of future VT. The area of severe wall thinning divided by the total area of thinning appears to be the best marker of risk.