Research

3.HemoVision: Unmasking SCTA’s Dynamic Blood Flow with Deep Learning

This  research project is focused on Scientific Machine Learning (SciML). the project integrates domain-specific models with cutting-edge artificial intelligence techniques. The work revolves around the amalgamation of structured scientific models, particularly differential equations, with the unstructured data-driven models inherent to machine learning. This synergy of scientific and data-driven approaches results in enhanced simulators that not only offer expedited computational performance but also provide more accurate approximations of complex systems.

Currently, Dr. Sangeeta’s is developing deep learning models for estimating Transient Hemodynamics from static subtraction computed tomographic angiography (SCTA) data. Typical CT angiography images do not contain any blood flow velocity or pressure information. However, the investigators believe that the raw unprocessed sinograms in sCTA imaging do contain spatio-temporal contrast transport information that is lost in image reconstruction (using filtered back projection or iterative reconstruction). Moreover, contrast transport is governed by the coupled mass transport (advection-diffusion equation) and blood flow (Navier-Stokes and mass conservation equations) physics. Consequently, the availability of time-stamped sCTA sinogram data will enable the estimation of the underlying dynamic blood flow velocity and relative pressure.

The major innovation in this project is the development of a novel label-free, physics-informed machine/deep learning-based image processing method for estimating time-resolved, in-vivo blood flow velocities and relative pressure directly from unprocessed time-stamped sCTA sinograms. Currently, hemodynamic analysis in vascular diseases is limited by the accuracy and reliability of state-of-the-art methods for estimating blood flow velocities and relative pressures. This project, for the first time, will enable the resolution of dynamic blood flow velocity and relative pressure maps from inherently static subtraction computed tomographic angiography (sCTA) data.

Pages: 1 2 3