kCSD-python, a tool for reliable current source density estimation

Kernel Current Source Density (kCSD), which we introduced in 2012, is a kernel-based method to estimate current source density (CSD) from extracellular potentials recorded with arbitrarily placed electrodes. Estimating reconstruction errors in CSD has been an outstanding challenge. To address it, here we revisit kCSD and explore its mathematical underpinnings. First, we quantify the information that can be recovered from extracellular recordings for a given setup, by introducing eigensources — a set of basic CSD profiles, which form the basis of estimation space. Next, we investigate the effect of relative placement of basis sources and electrodes on the reconstruction fidelity. We show that the correct distribution of sources is crucial for the reconstruction, in particular, CSD reconstruction is possible even for badly misplaced electrodes. We also introduce L-curve, a new method for choosing reconstruction parameters, in addition to the previously used cross-validation. Finally, we propose two types of diagnostics of reconstruction veracity, error propagation map and reliability map. For any given setup, the error propagation map indicates how the electrode noise propagates to the reconstructed CSD and the reliability map illustrates the point-wise reliability of kCSD estimation. The kCSD method and the additional techniques introduced here are implemented in kCSD-python, a new Python package provided under an open license. kCSD-python’s features and usage are highlighted with a jupyter notebook tutorial. This new tool can perform CSD estimations for 1D, 2D, and 3D electrode setups, assuming distributions of sources in a tissue, a slice, or in a single cell.

Abstract 18492: Phase Analysis Detects Human Atrial Fibrillation Sources While Classical Activation Mapping May Not: Reconciling Classical and Computational Mapping

Introduction: The mechanisms maintaining human persistent AF are elusive. It is striking how most optical mapping studies in animal and recently human AF show rotors and focal sources, while most classical activation mapping studies of electrograms do not. We tested the hypothesis that sites in human persistent AF showing rotors by phase analysis may, due to precession (‘wobble’) and fibrillatory collision, rarely reveal sources in activation maps. Methods: We studied 25 patients with persistent AF (LA 47 mm, CHADS2=1.9), in whom phase-mapping of electrograms from 64 pole baskets revealed rotors/focal sources where ablation terminated AF. Electrograms (fig A) were annotated (Matlab) using minimum dV/dt (unipoles, fig B) and peak amplitude criteria (bipoles) to create contours (isochrones), that were classified into a) complete, b) partial or c) unresolvable sources. Results: In each case, ablation at phase-identified rotors/sources (4.0±5.7 mins) terminated persistent AF to sinus rhythm (fig C, 64%) or atrial tachycardia. Notably, isochrones detected sources in only 5/25 (20%) of cases (fig D), more easily in unipolar than bipolar signals. Isochrones revealed partial sources in 11 (44%) and were unresolvable in 9 (36%). Source detection in classical maps was obscured by low signal: noise, varying sequence (rotor precession), or electrode noise that phase analysis resolved by analyzing neighboring sites (fig E). The figure summarizes these steps for a case with perfect agreement between activation and phase maps. Conclusions: Rotors and focal sources for human persistent AF detected by phase analysis were mostly undetected in activation maps, due to rotor precession and fibrillatory conduction. These data may inform approaches to revise classical criteria to better map AF.