The effect of catheter geometries on non-contact cardiac membrane potential mapping: an in-silico evaluation

Background

Non-paroxysmal atrial fibrillation remains challenging to treat effectively by catheter ablation. To improve therapeutic outcomes, diagnostic mapping techniques based on intracardiac electrode signals have been developed. However, conventional activation-based mapping methods are limited by insufficient spatial resolution, the need for electrode–tissue contact, and restricted field-of-view.

In our previous study, we proposed a deep-learning model, PointSenseNet (PSN), demonstrating the feasibility of high-resolution cardiac membrane potential mapping without electrode–tissue contact in simulation. Nevertheless, the influence of catheter geometries on mapping accuracy and field coverage remains unclear.

Objective

This study aims to perform an in-silico evaluation of the effect of field-of-view expansion on mapping accuracy using two representative catheter geometries: a two-dimensional radial-shaped and a three-dimensional basket-shaped configuration.

Method

We developed PSN, a deep learning model that predicts membrane potentials on cardiac tissue surfaces from point-cloud representations of tissue surface, electrode positions, and corresponding electrode signals.

Datasets were generated using a cardiac electrophysiology simulator based on the Courtemanche atrial model with regionally proliferated fibroblasts.

For the radial geometry, the non-contact distance ranged from 0–10 mm, with random tilt (0–5°) and each spine rotation (0–360°) to mimic clinical flexibility. The basket geometry used 0–15 mm non-contact distance, also with random tilt and full rotational variation. The radial design included five spines (12 mm each, four electrodes per spine), while the basket design had eight spines (eight electrodes each), of which 32 equatorial electrodes were used for recording (equatorial diameter 20 mm; tip-to-tip 15 mm).

Simulations were conducted on tissue areas of (i) 10 × 10, (ii) 20 × 20, and (iii) 30 × 30 mm² to evaluate the relationship between mapping accuracy and field-of-view. Model performance was assessed using mean absolute error (MAE).

Results

Mapping results (Figure A) showed that for the radial geometry, PSN achieved high-resolution reconstruction of complex excitation dynamics within 10 × 10 mm and 20 × 20 mm fields, even when spines were partially folded or overlapped. However, accuracy decreased markedly at larger 30 × 30 mm field size. In contrast, the basket geometry maintained robust mapping performance even at the largest field. Quantitative analysis (Figure B) showed that the 3D structure of the basket geometry suppressed accuracy degradation with increasing field size, achieving a mean error below 4 mV at the largest field.

Conclusions

This study demonstrates that the basket-shaped geometry enables high-resolution, wide-range membrane potential mapping under non-contact conditions using the PSN-based system.