Rapid prediction of cardiac activation in the left ventricle with geometric deep learning: a step towards cardiac resynchronization therapy planning
By
Ehsan Naghavi
Haifeng Wang
Vahid Ziaei-Rad
Julius Guccione
Ghassan Kassab
Vishnu Boddeti
Seungik Baek
Lik-Chuan Lee
February 7, 2026
Clinical Scorecard: Accelerated Assessment of Left Ventricular Cardiac Activation Using Geometric Deep Learning: Advancing Planning for Cardiac Resynchronization Therapy
At a Glance
Category Detail
Condition Dyssynchronous heart failure requiring cardiac resynchronization therapy (CRT) Key Mechanisms Use of geometric deep learning models (GNN and GINO) to predict left ventricular activation time maps for optimal pacing site identification Target Population Patients with dyssynchronous heart failure considered for CRT Care Setting Pre-procedural planning and optimization in cardiology clinical settings
Key Highlights
Approximately one-third of CRT recipients do not respond, partly due to suboptimal left ventricular lead placement. Geometric deep learning models (GNN and GINO) can predict LV activation times rapidly and accurately, aiding personalized CRT planning. An interactive web-based interface enables clinical decision support for individualized CRT optimization. Guideline-Based Recommendations
Diagnosis
Assess dyssynchronous heart failure patients for CRT candidacy based on clinical and imaging criteria. Evaluate left ventricular activation patterns to guide lead placement. Management
Optimize left ventricular lead placement to improve CRT response rates. Consider patient-specific anatomical variability when planning CRT. Utilize advanced computational models to identify optimal pacing sites pre-procedurally. Monitoring & Follow-up
Monitor clinical response and left ventricular remodeling post-CRT to assess efficacy. Adjust pacing settings or lead position if suboptimal response is observed. Risks
Non-response to CRT due to suboptimal lead placement. Potential procedural complications related to lead implantation. Patient & Prescribing Data
Patients with heart failure and ventricular dyssynchrony eligible for CRT
Personalized lead placement guided by geometric deep learning models may improve CRT response and reduce non-responder rates.
Clinical Best Practices
Incorporate patient-specific left ventricular geometry and activation data into CRT planning. Use computational modeling tools to simulate and select optimal pacing sites before implantation. Employ interactive decision-support interfaces to assist clinicians in individualized CRT optimization. References
