An Effective Data-Driven Diagnostic Strategy for Cardiac Pathology Screening
Abstract
In this research, we propose an effective data-driven diagnostic strategy to identify atrial fibrillation (AF) episodes. Published research so far has targeted AF detection through univariate and multivariate analysis of R-R interval.
As a potential enhancement, we suggested an advanced diagnostic methodology based on three dynamic patterns, namely the R-R interval and its first and second derivatives. Accordingly, we have targeted 11 metrics to describe each dynamic pattern, including four time-domain features, and seven non-linear features, yielding 33 detectors. Therefore, to conduct a suitable detection strategy for pathological AF screening, a dimensionality reduction process using a factor analysis technique is implemented to provide a homogeneous combination of the most relevant detectors with only 14 inputs. To demonstrate the effectiveness of the proposed approach, support vector classification algorithm trained on the 14-reduced-features has achieved, an average precision of 98.77% for validation and 98.78% for testing, calculated with 10-fold cross-validation.