Integrating standard-of-care clinical stroke workup within in silico embolic stroke models for etiology disambiguation

Presentation Number: TP388

Ricardo Roopnarinesingh*¹, Sreeparna Majee¹, Leon A. Rinkel², Jonathan M. Coutinho², Kelly Cao¹, Debanjan Mukherjee^1
1University of Colorado, Boulder, Boulder, Colorado, United States; ²Amsterdam University Medical Center, Amsterdam, Netherlands

Disclosures:
Ricardo Roopnarinesingh: No relevant financial relationships to disclose at the current time or during the last 24 months. | Debanjan Mukherjee: No relevant financial relationships to disclose at the current time or during the last 24 months. | Sreeparna Majee: No relevant financial relationships to disclose at the current time or during the last 24 months. | Leon A Rinkel: Jonathan M Coutinho: Consultant: Bayer (research support, all fees paid to my employer), Siemens (research support, all fees paid to my employer), AstraZeneca (research support, all fees paid to my employer) - Active (exists now) | Ownership Interest: TrianecT (co-founder and shareholder) - Active (exists now) Kelly Cao:

Abstract Body
Introduction: Embolic Stroke of Undetermined Source (ESUS) accounts for a critical proportion of all
ischemic strokes. Disambiguating embolism etiology is important to improve treatment efficacy and reduce
recurrent events. Patient-specific in silico models can shed valuable insights on embolus source-destination
mapping. This requires reliable and accurate pre/post-stroke hemodynamic models, which benefit from
integrating multiple modes of patient information from imaging and clinical records. This is a major state-
of-the-art challenge. Here, we present a workflow for multi-modal data integration from standard-of-care
workup towards recreating a data-rich digital twin of stroke patients.
Methods: Our workflow integrates non-contrast and contrast-enhanced head-neck CT and cardiac CT,
trans-thoracic echo, and perfusion imaging, along with clinical variables such as HR, systolic/diastolic vol-
umes, and stroke locations (with NIHSS scores). Quantitative data from these sources are then integrated
into a hemodynamic model by processing features such as arterial structure, inlet flow, tuned resistance
boundary conditions, cardiac timing, and stroke location. Resulting hemodynamic data was used to further
simulate embolus movement towards stroke site. Statistical sampling simulations using this model were
conducted to evaluate the likelihood that an occlusion location corresponded to cardiogenic, aortogenic, or
other arterial sources.
Results: We present our complete in silico workflow, and demonstrate the outcomes using a small cohort
of 5 patients acquired from a clinical database (anonymized, IRB exempt). We demonstrate that the workflow
yields high-resolution space-time varying patient hemodynamic patterns. Additionally, the embolus source-
destination likelihood mapping provides detailed quantitative insights on the embolism etiology in these
stroke patients. These findings indicate that our workflow and resulting digital twins can be a valuable tool
in addressing the current clinical challenges in discerning embolism etiology in ESUS cases.
Conclusions: We introduce a pipeline of transforming raw patient-specific information from multi-modal
imaging and clinical parameters into a cohesive, data-rich in silico model for embolic stroke comprising the
full heart-to-brain pathway. This offers a flexible digital twin approach for elucidating stroke etiologies in
patient-specific scenarios.