Anatomic Imaging Derived 4D Hemodynamics using Deep Learning

SUMMARY - ABSTRACT
Thoracic aortic aneurysm is a highly prevalent disease which can lead to devastating complications including
dissection or rupture. Early detection and regular monitoring of these patients via regular surveillance imaging is
essential to guide therapy management. The current paradigm for risk assessment in these patients is based on
primitive size thresholds with poor predictive value for aortic complications. There is strong evidence that 4D
hemodynamic biomarkers are drivers of aortic complications and can improve risk assessment and therapy
management. To obtain these biomarkers, a highly specialized MRI technique - 4D flow MRI - is required which
allows for the direct in vivo measurement of aorta 4D hemodynamics. However, several limitations impede wider
clinical translation, including the lack of access to dedicated MRI systems and 4D flow MRI protocols,
burdensome and time-consuming (30+ minutes) post-processing, and interpretation of 4D flow data requiring
dedicated software and highly specialized expertise.
To address these limitations, Third Coast Dynamics is developing a cloud-based artificial intelligence (AI)-based
platform called TCDflow that can replace 4D flow MRI by providing 4D hemodynamic output directly from widely
available routine and easy-to-obtain clinical anatomic images of the chest. Our proof-of-concept studies have
leveraged a large database of >6700 existing 4D flow MRI patient data to develop a prototype TCDflow fluid
physics informed deep learning neural network for the prediction of 4D aortic hemodynamics using anatomic
images as input data. Further development, evidence generation, and steps toward commercialization will be
conducted in a two-phase approach. Phase 1 (P1) focuses on further development and fine-tuning of the This
Coast Dynamics analysis pipeline (P1, Aim 1). The technology will then be tested in a large, single center
(Northwestern) retrospective aorta outcomes study (P1, Aim 2). These developments and validation will provide
the foundation for Phase 2 which focuses on developing our clinician-facing cloud-based analysis platform and
report (P2, Aim 1), performing a large multicenter retrospective validation and outcomes study (P2, Aim 2),
completing an end-user TCDflow evaluation (P2, Aim 3), and securing FDA 510(k) clearance (P2, Aim 4). The
completion of the Phase 1 and 2 deliverables will result in an FDA-cleared product which can be readily
commercialized. All aims are designed with guidance from consultants with direct expertise in FDA 510(k)
clearance of digital health products.
The technology will provide improved personalized risk-stratification of aortic complications beyond the current
simple and insufficient clinical measures. Increasing operational efficiencies and reduced health care utilization
costs will be achieved by access to cloud-based 4D hemodynamic assessment by a wide range of patients and
healthcare providers without the need for highly specialized imaging equipment, training, and expertise.

Grant Number: 4R42HL174259-02
NIH Institute/Center: NIH
Principal Investigator: Bradley Allen