Published: 2026-02-07 20:06

AI Predicts Cardiac Activation for Resynchronization Therapy Planning

Cardiac resynchronization therapy (CRT) is a cornerstone treatment for selected patients with heart failure and electrical dyssynchrony, aiming to improve cardiac function and patient outcomes. However, identifying the optimal pacing site within the left ventricle to achieve maximum benefit remains a complex challenge. New research, published in *npj Digital Medicine*, explores the potential of geometric deep learning to rapidly predict cardiac activation patterns, offering a promising step towards more precise CRT planning.

Understanding Cardiac Resynchronization Therapy (CRT)

CRT involves implanting a specialised pacemaker that stimulates both ventricles, and often the right atrium, to restore a more coordinated contraction. It is primarily indicated for patients with symptomatic heart failure, reduced left ventricular ejection fraction (LVEF), and evidence of electrical dyssynchrony, typically a wide QRS complex or left bundle branch block (LBBB).

The goal of CRT is to resynchronise the contraction of the left ventricle, which can be asynchronous in heart failure due to conduction abnormalities. By pacing specific areas, particularly in the left ventricle, clinicians aim to:

• Improve left ventricular systolic function.
• Reduce mitral regurgitation.
• Enhance exercise capacity.
• Alleviate heart failure symptoms.
• Improve quality of life and survival.

Despite its proven benefits, a significant proportion of patients (around 30-40%) do not respond optimally to CRT. This variability in response is often attributed to the difficulty in precisely identifying the most effective pacing location, which can vary considerably between individuals.

The Challenge of Optimal Pacing Site Selection

Current clinical practice for CRT lead placement relies on a combination of imaging modalities and electrophysiological mapping. Techniques such as echocardiography, cardiac MRI, and CT venography help identify suitable coronary sinus branches for lead placement and assess myocardial viability. Intracardiac electrophysiological mapping during the implant procedure provides real-time information on local activation times and tissue characteristics.

However, these methods have limitations:

• Time-consuming: Detailed mapping can prolong procedure times.
• Invasive: Electrophysiological mapping requires additional catheters and manipulation within the heart.
• Variability: The optimal site can be difficult to pinpoint definitively, leading to suboptimal lead placement in some cases.
• Static assessment: Many current assessments provide a snapshot, not a dynamic prediction of how pacing at different sites will alter activation.

The ideal scenario involves placing the left ventricular lead at a site of late electrical activation within viable myocardium, avoiding scar tissue. Achieving this consistently and efficiently is a major focus of ongoing research and technological development.

Geometric Deep Learning: A New Approach

The research highlights the application of geometric deep learning to predict cardiac activation in the left ventricle. Geometric deep learning is a subfield of artificial intelligence that extends deep learning techniques to handle data structured in non-Euclidean spaces, such as graphs or meshes. In cardiology, this is particularly relevant for analysing complex anatomical structures like the heart, which can be represented as 3D meshes.

The study leveraged this advanced AI methodology to:

• Analyse detailed anatomical and electrophysiological data from patient hearts.
• Learn the intricate relationships between ventricular geometry, myocardial properties, and electrical activation patterns.
• Rapidly predict how electrical impulses would propagate through the left ventricle from various potential pacing sites.

The “rapid prediction” aspect is key. Traditional computational models of cardiac electrophysiology are often complex and computationally intensive, making them impractical for real-time clinical decision-making during an implant procedure. By contrast, a well-trained deep learning model can provide predictions almost instantaneously.

How AI Could Transform CRT Planning

The potential implications of this AI-driven approach for CRT planning are significant, offering a pathway towards more personalised and effective therapy:

Enhanced Pre-Procedural Planning

Before a patient undergoes CRT implantation, clinicians could use AI models to simulate the effects of pacing at numerous potential sites. This would allow for:

• Optimised Lead Placement: Identifying the most effective target regions for lead implantation, potentially improving response rates.
• Reduced Procedure Time: By narrowing down the optimal sites pre-operatively, the need for extensive intra-procedural mapping might be reduced.
• Personalised Strategies: Tailoring the pacing strategy to each patient’s unique cardiac anatomy and electrophysiology, moving beyond a “one-size-fits-all” approach.

Improved Patient Outcomes

A more precise lead placement, guided by AI predictions, could lead to:

• Higher Response Rates: Increasing the proportion of patients who benefit significantly from CRT.
• Better Clinical Efficacy: Maximising improvements in LVEF, symptoms, and quality of life.
• Reduced Re-interventions: Minimising the need for lead repositioning or device adjustments post-implantation due to suboptimal initial placement.

Resource Utilisation in the NHS

For the UK healthcare system, the efficiency gains could be substantial. Shorter procedure times, fewer re-interventions, and improved patient outcomes could translate into:

• Cost Savings: Reducing the overall cost burden associated with CRT.
• Increased Capacity: Freeing up cath lab time and resources, potentially allowing more patients to access this life-saving therapy.
• Streamlined Workflows: Integrating AI tools into existing clinical pathways to enhance decision-making.

Clinical Implications and Future Outlook

While this research represents a significant advance, it is important to contextualise it as a foundational step. The findings demonstrate the feasibility and potential of geometric deep learning in this domain, but further rigorous validation is essential before clinical translation.

Key areas for future development include:

• Larger Cohort Studies: Testing the AI model on diverse and larger patient populations to confirm its generalisability and robustness.
• Prospective Clinical Trials: Evaluating the impact of AI-guided CRT planning on actual patient outcomes in controlled settings.
• Integration with Existing Tools: Developing user-friendly interfaces that allow seamless integration of AI predictions into current imaging and mapping systems.
• Regulatory Approval: Navigating the necessary regulatory pathways for medical devices and AI algorithms in clinical use.

The role of AI in clinical decision-making is to augment, not replace, the expertise of healthcare professionals. These tools are designed to provide clinicians with more precise, data-driven insights, enabling them to make more informed decisions and ultimately deliver better patient care. The rapid prediction of cardiac activation using geometric deep learning holds considerable promise for refining CRT planning and improving outcomes for patients living with heart failure.

Source: Nature