In-Procedure Personalized Atrial Digital Twin to Predict Outcome of Atrial Fibrillation Ablation

心房颤动（AF）是一种心律紊乱，主要影响老年人，增加中风风险并降低生活质量。房颤可以通过射频消融治疗，射频消融会不可逆地破坏引发和维持心律失常的组织区域。目前，消融成功与否只能在治疗完成后进行评估。心脏数字双胞胎提供了基于物理的患者表示，随着更多信息的可用而更新。这些双胞胎可用于预测患者对AF消融的反应。我们建议使用接受AF消融术的患者的数字双胞胎，从术前成像和重要的是，详细的术中侵入性测量中创建，以预测心房对消融的反应。这将为创建数字孪生模型提供基础，可用于指导和优化个体患者的房颤消融治疗。我们和其他人已经为个体患者创建了心脏模型，但个性化过程是时间密集型的，因为模型有许多自由度。因此，该项目的主要挑战是加快个性化和预测，以便在临床程序中可用的30-45分钟时间窗口内实现。我们将通过以下方式解决这一挑战：（1）优化使用程序前收集的个体患者信息，以及来自既往患者群体的信息;（2）识别关于个体信息最多的模型参数，并且可以从临床测量中识别;（3）选择有效的方法来加速计算模拟;以及（4）对快速过程中个性化的新方法进行系统研究，包括最先进的机器学习方法。整个过程将采用概率方法，预测将取决于数据和模型拟合的质量。然后，我们的工作流程将被整合，以生成一个易于使用的软件平台，该平台适合在临床环境中部署，用于20名患者的概念验证临床研究。该项目将开发技术，生成试点数据，创建商业案例，并与患者接触，以便我们可以进行单独和广泛的临床研究，以确定我们工作流程的有效性及其对常规临床使用的适用性。

Atrial fibrillation (AF) is a disorder of heart rhythm that predominantly affects older people, increasing the risk of stroke and reducing quality of life. AF can be treated by radio-frequency ablation, which irreversibly destroys regions of tissue that initiate and sustain the arrhythmia. Currently, success of the ablation can only be evaluated after the therapy has been delivered. Cardiac digital twins provide a physics-based representation of the patient that is updated as more information becomes available. These twins can be used to predict how a patient will respond to an AF ablation. We propose to use digital twins of patients undergoing an AF ablation procedure, created from both pre-procedure imaging, and importantly, detailed in-procedure invasive measurements, to predict how the atria will respond to ablation. This will provide the basis for creating digital twins that can be used to guide and optimise AF ablation therapies for individual patients. Heart models for individual patients have been created by us and others, but the personalisation process is time intensive because the models have many degrees of freedom. The main challenge for this project is therefore to accelerate both personalisation and prediction so that they can be achieved within the 30-45 minute time window that is available during a clinical procedure.We will address this challenge by (1) optimising the use of information about an individual patient that can be gathered prior to the procedure, as well as information from a population of previous patients; (2) identifying model parameters that are most informative about an individual and which can be identified from clinical measurements; (3) selecting effective methods to accelerate computational simulations; and (4) a systematic investigation of novel approaches for rapid in-procedure personalisation, including state-of-the-art machine learning methods. A probabilistic approach will be taken throughout, and predictions will depend on the quality of the data and model fits. Our workflow will then be integrated to produce an easy to use software platform that is suitable for deployment in the clinical setting for a proof-of-concept clinical study in 20 patients. The project will develop the technology, generate pilot data, create a business case, and engage with patients so that we can undertake a separate and extensive clinical study to establish the effectiveness of our workflow and its suitability for routine clinical use.