Machine-Learning-Based Multi-Modal Force Estimation for Steerable Ablation Catheters.

Area of Science:

• Cardiovascular Medicine
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Catheter-based cardiac ablation is crucial for treating atrial fibrillation (AF).
• Accurate contact force measurement between the catheter and heart tissue is vital for effective lesion creation and procedural success.
• Current methods may lack precision, necessitating improved sensing technologies.

Purpose of the Study:

• To develop and validate a novel multi-modal contact force estimator for catheter-based cardiac ablation.
• To leverage deep learning, specifically Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), for enhanced force estimation.
• To assess the estimator's performance using simulated fluoroscopy and real-world data acquisition.

Main Methods:

• A multi-modal approach using catheter shape and optical flow from video data.
• Feature extraction via transfer learning from image and optical flow modalities.
• Integration of Long Short-Term Memory Networks (LSTMs) with a memory fusion network (MFN) to capture temporal dependencies and hysteresis.
• Spatial and temporal network architecture with late fusion techniques (concatenation of LSTMs, transformer decoders, GRUs).

Main Results:

• The proposed multi-modal network achieved a mean absolute error of only 2.84% of the total magnitude.
• Performance was superior to single-modality networks and basic late fusion concatenation.
• Data collected under realistic conditions demonstrated the estimator's practicality.

Conclusions:

• The developed multi-modal AI-based contact force estimator is highly accurate and practical for cardiac ablation.
• This sensor-free approach offers a significant improvement over existing methods.
• The findings highlight the potential of multimodal deep learning in advancing minimally invasive cardiac procedures.