AI-driven View Guidance System in Intra-cardiac Echocardiography Imaging

TL;DR

The paper tackles the challenge of operator-dependent ICE navigation by introducing an AI-driven, closed-loop view guidance system that estimates relative catheter joint states to move from any ICE view to predefined clinical views. It leverages a MedMamba-based regression model with a feature-mixing layer and quantile loss to map ICE images to a four-DoF joint state in SE(3), operating in a human-in-the-loop framework to update guidance continuously. Trained on a CARTO-derived dataset of 858 subjects and 143k training cases, the approach demonstrates 89% success in guiding views over 6,532 test cases and shows feasible human-in-the-loop validation with re-planning improving navigation accuracy. The results suggest that ICE view guidance can be robustly integrated into clinical workflows, potentially reducing procedure times and improving consistency across operators; future work includes beating-heart phantom validation and extended in vivo testing.

Abstract

Intra-cardiac echocardiography (ICE) is a crucial imaging modality used in electrophysiology (EP) and structural heart disease (SHD) interventions, providing realtime, high-resolution views from within the heart. Despite its advantages, effective manipulation of the ICE catheter requires significant expertise, which can lead to inconsistent outcomes, especially among less experienced operators. To address this challenge, we propose an AIdriven view guidance system that operates in a continuous closed-loop with human-in-the-loop feedback, designed to assist users in navigating ICE imaging without requiring specialized knowledge. Specifically, our method models the relative position and orientation vectors between arbitrary views and clinically defined ICE views in a spatial coordinate system. It guides users on how to manipulate the ICE catheter to transition from the current view to the desired view over time. By operating in a closedloop configuration, the system continuously predicts and updates the necessary catheter manipulations, ensuring seamless integration into existing clinical workflows. The effectiveness of the proposed system is demonstrated through a simulation-based performance evaluation using real clinical data, achieving an 89% success rate with 6,532 test cases. Additionally, a semi-simulation experiment with human-in-the-loop testing validated the feasibility of continuous yet discrete guidance. These results underscore the potential of the proposed method to enhance the accuracy and efficiency of ICE imaging procedures.

Figures (11)

• Figure 1: Overview of the proposed view guidance system. When users want to navigate to clinically predefined views during procedures, our proposed system provides continuous guidance on how to manipulate the ICE catheter via an interactive device ( e.g., a touchpad for view selection and a feedback monitor) until the target view is achieved. (ACUSON Origin Cardiovascular Ultrasound System and AcuNav Lumos ICE catheter, Source: Siemens Healthineers)
• Figure 2: Process of the proposed method. (a)Procedure Overview: The clinician first positions the ICE catheter at the home view and then selects the desired viewpoint. The proposed network provides the relative state, which is then transformed into robotic coordinates using inverse kinematics (IK). Guided by this information, the clinician manipulates the ICE catheter, and the procedure is performed iteratively. (b) Iteration step: During the procedure, the user can sequentially move toward the guided target state. The proposed method continuously updates the current state based on the home view state, enabling progress tracking from the starting point to the target point. Each step can be repeated as needed until the target view is achieved.
• Figure 3: (a) The reconstructed cardiac anatomical volume mesh of a single subject, generated using our volume contouring algorithm liao2018more. (b) The reconstructed volume mesh for missing structures such as the RA, RV, and LV, where the mesh centers were predicted based on a small existing dataset. (c) The composition of the CARTO dataset for a single subject, which includes multiple image views visualizing different structures in various positions, along with corresponding position/orientation data and the volume mesh. (d) Our target dataset consists of six target states, each corresponding to one of six clinically defined views for a single subject. While each target state shares the same position, the orientations differ, with the fan direction aligned to the center of the target volume.
• Figure 4: The proposed method architecture. The center of the block represents the MedMamba structure. We have added layers for feature mixing with target class code and image. At the output of the network, we added multi-head structure to estimate position and orientation separately. The details of SS2D structure are shown at the bottom. The image feature is split into sequences in different orders and put into the SSM model. Then, the outputs are merged together.
• Figure 5: Validation method. (a) Percentage of correct views calculation: The prediction is considered correct if the predicted view matches the target volume, and incorrect otherwise. (b) Distance calculation: The distance is measured from the center of the target volume to the predicted view.