Digital-twin-based Online Parameter Personalization for Implantable Cardiac Defibrillators.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Implantable cardioverter defibrillators (ICDs) require personalized parameter settings for optimal patient therapy.
• Current clinical guidelines lack specific recommendations for parameter personalization, particularly for complex or rare cardiac conditions.
• Evolving patient conditions necessitate dynamic adjustments to ICD therapies.

Purpose of the Study:

• To propose a reinforcement learning (RL) framework for online personalization of ICD parameter settings.
• To develop a method for creating a patient-specific digital twin using ECG signals for RL environment.
• To evaluate the performance of RL-based parameter personalization against default settings.

Main Methods:

• Utilized ECG signals from ECG patches to infer patient heart states.
• Developed a digital twin of the patient to serve as the environment for RL.
• Employed RL algorithms to explore and identify optimal ICD parameter settings.
• Conducted experiments on three virtual patients with diverse and evolving cardiac conditions.

Main Results:

• The proposed RL framework successfully identified personalized ICD parameter settings.
• The personalized settings demonstrated improved performance compared to default parameter settings in virtual patient models.
• The approach showed potential for reducing misdiagnosis by adapting to evolving heart conditions.

Conclusions:

• Reinforcement learning offers a viable framework for online ICD parameter personalization.
• Digital twin technology, combined with RL, can facilitate adaptive and individualized device therapies.
• This approach holds clinical relevance for patients with ICDs and ECG patches, enabling tailored adjustments to their device settings.