Maskless 2D/3D artificial subtraction angiography
AssigneesNorthwestern University, Ann & Robert H. Lurie Children's Hospital of Chicago
InventorsDonald R. Cantrell, Sameer A. Ansari, Leon Cho
Life of this patent
Estimating the term…
Abstract
During catheter-based angiography, the bone and soft tissues degrade visualization of the vasculature, which is of primary interest in such medical imaging procedures. The present disclosure includes systems and methods utilizing a trained neural network to remove the bone and soft tissue densities from post-contrast images, revealing isolated vascular densities, without the need for a standard pre-injection digital mask and in the setting of patient motion. The final angiographic images may be created in real-time. Systems and methods for the training and optimization of the disclosed neural network are also described.
Claims18
A method of generating a maskless artificial subtraction angiography image using a machine learning model comprising:
- obtaining raw angiographic images from an individual, wherein the raw angiographic images are three-dimensional, spatiotemporal, multi-frame angiographic inputs;
- utilizing spatial and temporal information in the raw angiographic images to identify vascular densities;
- and
- generating a final three-dimensional spatiotemporal angiographic image, wherein bone and soft tissue structures are removed from the final angiographic image, leaving isolated vascular densities,
- wherein a method for training and optimizing the method of generating the maskless artificial subtraction angiography image using the machine learning model further comprises the steps of:
- obtaining a database of raw angiographic images from a plurality of individuals, wherein the raw angiographic images are three-dimensional, spatiotemporal, multi-frame angiographic inputs;
- calculating motion estimates using the database of raw angiographic images;
- isolating motionless angiographic images from the database;
- transforming the motionless raw angiographic images into digital subtraction angiography images to generate input-output data pairs;
- applying the motion estimates to the input-output data pairs;
- and
- utilizing the data pairs for training and optimizing the generation of the maskless artificial subtraction angiography image.
A system for generating a maskless artificial subtraction angiography image comprising:
- an angiographic imaging device configured to generate raw angiographic images from an individual;
- a machine learning model configured to remove bone and soft tissue densities and isolate vascular densities, and wherein the machine learning model is further configured to compute motion estimates, and wherein the motion estimates are generated from a motion-degraded dataset;
- and
- a processing device configured to generate a final angiographic image in real-time or near real-time, wherein bone structures are removed from the final angiographic image.
A method of angiographic motion stabilization in artificial subtraction angiography comprising:
- collecting a plurality of raw angiographic images, wherein the raw angiographic images are three-dimensional, spatiotemporal, and multi-frame angiographic inputs;
- matching known anatomical features from the plurality of raw angiographic images;
- computing interframe motion from matched anatomical features;
- computing a motion-stabilized series of angiographic images using the computed interframe motion;
- and
- generating a digital subtraction angiography from a motion-stabilized dataset, wherein bone structures are removed from a final angiographic image.
Dependencies are inferred from the claim text.