Pose Estimation for Intra-cardiac Echocardiography Catheter via AI-Based Anatomical Understanding

TL;DR

Problem: ICE catheter localization traditionally relies on EM tracking, which is susceptible to magnetic interference and drift or requires operator-driven view adjustments. Approach: An anatomy-aware pose estimation system derives the catheter's position $P$ and orientation $O$ solely from ICE images using a Vision Transformer, with a [CLS] token feeding two linear heads and a composite loss $l_{total}$ where $l_{total} = l_{mse}(\hat{P}, P) + \lambda \, l_{mse}(\hat{O}, O)$ and $\lambda = 2$. Key findings: On data from 851 subjects, the method yields an average positional error of 9.48 mm and orientation errors of $(16.13, 8.98, 10.47)$ degrees, with qualitative 3D mesh alignment of predicted and target views. Significance: Enables tracking-free ICE navigation and can complement systems like CARTO, reducing operator workload and providing real-time anatomical localization during ICE-guided interventions.

Abstract

Intra-cardiac Echocardiography (ICE) plays a crucial role in Electrophysiology (EP) and Structural Heart Disease (SHD) interventions by providing high-resolution, real-time imaging of cardiac structures. However, existing navigation methods rely on electromagnetic (EM) tracking, which is susceptible to interference and position drift, or require manual adjustments based on operator expertise. To overcome these limitations, we propose a novel anatomy-aware pose estimation system that determines the ICE catheter position and orientation solely from ICE images, eliminating the need for external tracking sensors. Our approach leverages a Vision Transformer (ViT)-based deep learning model, which captures spatial relationships between ICE images and anatomical structures. The model is trained on a clinically acquired dataset of 851 subjects, including ICE images paired with position and orientation labels normalized to the left atrium (LA) mesh. ICE images are patchified into 16x16 embeddings and processed through a transformer network, where a [CLS] token independently predicts position and orientation via separate linear layers. The model is optimized using a Mean Squared Error (MSE) loss function, balancing positional and orientational accuracy. Experimental results demonstrate an average positional error of 9.48 mm and orientation errors of (16.13 deg, 8.98 deg, 10.47 deg) across x, y, and z axes, confirming the model accuracy. Qualitative assessments further validate alignment between predicted and target views within 3D cardiac meshes. This AI-driven system enhances procedural efficiency, reduces operator workload, and enables real-time ICE catheter localization for tracking-free procedures. The proposed method can function independently or complement existing mapping systems like CARTO, offering a transformative approach to ICE-guided interventions.

Figures (3)

• Figure 1: Each colored dot represents the left atrium and pulmonary vein structure of a single subject. The gray fan indicates the current ICE image view, serving as the input to our model. The red, blue, and green arrows within the purple box illustrate the predicted orientation orientation and position of the imaging plane, which is expressed in anatomy-relative coordinates.
• Figure 2: Network architecture: The input ICE image is patchified and fed into the ViT. The [CLS] token from the output is then passed through separate linear layers to estimate position and orientation independently.
• Figure 3: Representative results of the proposed method: The top-left case visualizes the LAA view, the top-right case visualizes the LPV view, the bottom-left case visualizes the ESO view, and the bottom-right case visualizes the RPV view.