[Paper Review] AI-powered multimodal modeling of personalized hemodynamics in aortic stenosis

Çağlar Öztürk, Daniel H. Pak+7 more|arXiv (Cornell University)|Jun 29, 2024

Cardiac Valve Diseases and TreatmentsMedicine3 citations

TL;DR

This paper presents an AI-powered, fully automated framework for high-fidelity, patient-specific modeling of aortic stenosis hemodynamics using only CT angiography. By leveraging deep learning for rapid and accurate meshing, the method enables fast, reliable fluid-structure interaction simulations and physical benchtop models that accurately reproduce clinical Doppler hemodynamic measurements, significantly accelerating personalized treatment planning and device optimization.

ABSTRACT

Aortic stenosis (AS) is the most common valvular heart disease in developed countries. High-fidelity preclinical models can improve AS management by enabling therapeutic innovation, early diagnosis, and tailored treatment planning. However, their use is currently limited by complex workflows necessitating lengthy expert-driven manual operations. Here, we propose an AI-powered computational framework for accelerated and democratized patient-specific modeling of AS hemodynamics from computed tomography. First, we demonstrate that our automated meshing algorithms can generate task-ready geometries for both computational and benchtop simulations with higher accuracy and 100 times faster than existing approaches. Then, we show that our approach can be integrated with fluid-structure interaction and soft robotics models to accurately recapitulate a broad spectrum of clinical hemodynamic measurements of diverse AS patients. The efficiency and reliability of these algorithms make them an ideal complementary tool for personalized high-fidelity modeling of AS biomechanics, hemodynamics, and treatment planning.

Motivation & Objective

• To overcome the bottleneck of manual, time-consuming geometric reconstruction in patient-specific modeling of aortic stenosis.
• To develop an AI-powered, fully automated meshing pipeline that accelerates and democratizes high-fidelity hemodynamic modeling.
• To integrate automated geometry with fluid-structure interaction and soft robotics models for accurate recapitulation of clinical hemodynamic measurements.
• To enable both in silico and in vitro patient-specific modeling with high temporal and spatial resolution using only CT data.
• To support personalized treatment planning and device optimization by accurately simulating complex hemodynamics such as flow vorticity and stasis.

Proposed method

• Employed deep learning-based algorithms to automate 3D mesh generation from CT angiography, achieving 100× faster processing than conventional methods.
• Used AI-driven meshing to produce task-ready geometries for both computational fluid dynamics and physical benchtop simulations.
• Integrated fluid-structure interaction (FSI) models to simulate dynamic valve motion and blood flow at high temporal resolution.
• Developed patient-specific soft robotic left ventricular sleeves using 3D printing and thermoplastic polyurethane to mimic physiological contraction.
• Constructed a hydrodynamic flow loop with adjustable resistance and compliance chambers to replicate patient-specific hemodynamic conditions.
• Validated results using continuous-wave and color flow mapping Doppler, comparing simulated data to clinical echocardiographic measurements.

Experimental results

Research questions

• RQ1 Can AI-powered automated meshing significantly reduce the time and effort required for patient-specific aortic stenosis modeling?
• RQ2 Can the proposed framework accurately reproduce clinical Doppler hemodynamic measurements using only CT data?
• RQ3 To what extent can the integration of FSI and soft robotics models replicate real patient hemodynamics in both in silico and in vitro settings?
• RQ4 How does the accuracy and speed of the AI-driven meshing approach compare to traditional manual reconstruction techniques?
• RQ5 Can this framework support personalized treatment planning by simulating complex hemodynamic features like flow vorticity and stasis?

Key findings

• The AI-powered meshing pipeline achieved 100× faster geometry reconstruction compared to conventional manual methods while maintaining higher accuracy.
• The framework successfully generated patient-specific 3D models from CT data that enabled accurate simulation of hemodynamic parameters such as peak velocity and mean pressure gradient.
• In silico simulations using fluid-structure interaction models accurately recapitulated clinical continuous-wave Doppler measurements of aortic stenosis patients.
• The physical benchtop models, combining 3D-printed anatomies with soft robotic LV sleeves, reproduced patient-specific hemodynamics with high fidelity, validated by Doppler echocardiography.
• Dimensionless velocity index (DVI) calculated from simulations closely matched clinical values, confirming the model's physiological relevance.
• The integration of AI-driven meshing with multimodal modeling enables scalable, reproducible, and clinically relevant patient-specific hemodynamic analysis.