Prediction of CRT Response Using a Lead Placement Score Derived from 4DCT

Abstract

AbstractBackgroundCardiac resynchronization therapy (CRT) is an effective treatment for patients with heart failure; however, 30% of patients do not respond to the treatment. We sought to derive patient-specific left-ventricle (LV) maps of lead placement scores (LPS) that highlight target pacing lead sites for achieving a higher probability of CRT response.MethodsEighty-two subjects recruited for the ImagingCRT trial were retrospectively analyzed. All 82 subjects had two contrast-enhanced full-cardiac cycle 4DCT scans: a baseline and a 6-month follow-up scan. CRT response was defined as a reduction in CT-derived end-systolic volume ≥15%. Eight LV features derived from the baseline scans were used to train a support vector machine (SVM) via a bagging approach. An LPS map over the LV was created for each subject as a linear combination of the SVM feature weights and the subject’s own feature vector. Performance for distinguishing responders was performed on the original 82 subjects.ResultsFifty-two (63%) subjects were responders. Subjects with an LPS≤Q1 (lower-quartile) had a posttest probability of responding of 14% (3/21), while subjects with an LPS≥ Q3 (upper-quartile) had a posttest probability of responding of 90% (19/21). Subjects with Q1<LPS<Q3 had a posttest probability of responding that was essentially unchanged from the pretest probability (75% vs 63%, p=0.2). An LPS threshold that maximized the geometric mean of true-negative and true-positive rates identified 26/30 of the non-responders. The AUC of the ROC curve for identifying responders with an LPS threshold was 87%.ConclusionsAn LPS map was defined using 4DCT-derived features of LV mechanics. The LPS correlated with CRT response, reclassifying 25% of the subjects into low-probability of response, 25% into high-probability of response, and 50% unchanged. These encouraging results highlight the potential utility of 4DCT in guiding patient selection for CRT. The present findings need verification in larger independent data sets and prospective trials.Clinical PerspectiveCardiac resynchronization therapy (CRT) is a proven treatment for patients with heart failure and dyssynchrony; however, approximately 30% of patients do not respond to the treatment. Additionally, the relatively high non-responder rate poses difficulties for the optimal utilization of medical resources; thus, more accurate patient stratification for CRT remains an unmet need. Despite significant efforts focused on using imaging to guide CRT, the results thus far have been ambiguous. Poor reproducibility of echocardiography coupled with the complexity of cardiac magnetic resonance have likely contributed to the poor overall adoption of these methods for pre-CRT assessment. In this work, we describe a metric called the lead placement score (LPS) that combines multiple 4DCT-derived features of left-ventricular (LV) mechanics into a single number for each possible pacing lead location on the LV; the features included in the LPS map have previously been shown to correlate with CRT response. Using a machine learning classifier, a model was constructed with these features and then used to derive the LPS map for each individual subject. The LPS was found to correlate with the probability of a subject responding to CRT. 4DCT is widely available and provides high-resolution images of the full cardiac cycle. Additionally, recent technological advancements have also dramatically reduced the radiation dose from 4DCT scans. The advantages of 4DCT coupled with the promising results reported in this study, highlight the potential utility of 4DCT in the planning of CRT.