Utilization of regression analysis for conductance catheters: considerable improvement in volume estimation

Error in a simulated conductance catheter volume system was reduced over 60% by statistically incorporating a priori information through training. The conventional deterministic model uses a specific cylindrical geometry but does not otherwise incorporate the relationship between potentials and volume. The statistical model uses a weighted sum of inverse differences of potentials, the weights determined through training. With a 288 point set of volume/potential training data a mean absolute root-mean-square error of 2.44% was obtained, while the conventional method gave 17.5% error. The conventional method underestimated the volume and exhibited an offset due to violations of the cylindrical model assumptions. The regression method yielded a substantial improvement in performance