Dynamic reconstruction and rendering of 3d tomosynthesis images

The method includes the steps of loading a set of projection images into a memory device, determining a reconstruction method for the set of projection images, reconstructing a 3D tomographic image from the set of projection images to be displayed to a user; and performing any post reconstruction processing on the 3D tomographic image.

Dynamic reconstruction and rendering of 3d tomosynthesis images

Dynamic tomographic image reconstruction and rendering on-demand - Real-Time Tomography, LLC

In this example, the x-ray focus is moved continuously while a small number of discrete linear detectors produce a sequential set of linear images of the anatomy. Note that only a small number of acquisition locations are shown for clarity.

This geometry is similar to that described in Patent Application Publication No. Regardless of acquisition geometry used, after the projection images are acquired, they are reconstructed into a set of 3D tomographic images that are saved and then reviewed at a later time. In stepall 3D tomographic images are reconstructed at fixed increments e.

Once a 3D image dataset is reconstructed, the images are saved at step Once saved, the process proceeds to step whereby a user e. If the user wishes to change any of the reconstruction parameters used to process the images, the process may begin again and a new set of 3D tomographic images must be reconstructed.

This reconstruction method is essentially an off-line approach. After the projection images are acquired, they must be reconstructed into 3D tomographic images.

Iterative tomosynthesis reconstruction methods, such as algebraic reconstruction techniques and maximum likelihood estimation maximization, generally involve reconstructing a 3D tomographic image of the full imaged volume.

These techniques provide good image quality but are computationally expensive as they generally involve an initial reconstruction followed by iterative updates of the full 3D image dataset until a threshold criterion is met.

In DBT, typical reconstruction times vary from 5 to 30 minutes to reconstruct one 3D image dataset per breast. Single-pass reconstruction techniques used in tomosynthesis do not generally require a reconstruction of the full 3D image dataset and may thus be computationally efficient.

Dynamic tomographic image reconstruction and rendering on-demand - Real-Time Tomography, LLC

Single-pass reconstruction allows reconstruction of a 3D tomographic image in a single iteration consisting of a set of reconstruction steps.

In this example, a set of reconstruction steps is defined as the performance of steps and as these steps may be iteratively performed numerous times, each time resulting in the reconstruction of a single image. Examples of single-pass reconstruction techniques are shift-and-add algorithms and simple backprojection in which 2D projection images are spatially translated with respect to each other through the image plane hence the name "backprojection" to obtain a rough approximation of the original.

Dynamic reconstruction and rendering of 3d tomosynthesis images

The projections interact constructively in regions that correspond to the structures in the original object. Structures not in the reconstructed image plane are blurred.

Filtered backprojection FBP is another backprojection method in which the projection images are filtered prior to reconstruction. Each of these conventional reconstruction techniques has inherent drawbacks, e. By maintaining a large number of image files for reconstruction, each technique is also highly demanding on a central processing unit CPUthereby causing additional processing delays.

Several approaches have been attempted to overcome these drawbacks, such as application specific integrated circuits ASICs of specifically designed field programmable gate arrays FPGAswhich is a device that can be configured to perform a specific task.

Both these approaches are expensive though, as the ASICs or FPGAs are designed for a single, specific purpose and do not provide much, if any, additional scalability.

Graphic Processor Units GPUs are pipeline processors, designed to accelerate the graphics rendering pipeline. Many of the gains in GPU performance have arisen from the ability to parallelize the various elements of the pipeline. Traditionally, the GPU architecture included vertex processors, called vertex shaders, which were specialized for geometric computations, and pixel processors or pixel shaders which were specialized for point operations.

More recently, GPUs have been based upon a unified shader architecture in which unified processors will switch between the two types of shaders depending on the work that needs to be done.

Graphics objects are typically composed of polygon meshes, where additional surface detail can be modeled by mapping images or textures onto the polygons during the rendering phase.

Texture mapping is a technique for efficiently modeling a surface's properties and is an efficient way to provide intricate surface detail without increasing an object's polygon count.

GPUs are highly optimized to perform texture mapping very quickly, even under perspective distortion. The input to the pipeline is a list of geometric objects specified by vertices and temporarily stored in a vertex buffer A vertex is the point intersection of two sides of a polygon.

The vertices are transformed to the object's position in 3D space and projected to the screen plane by vertex shaders The projected vertices are then assembled into triangles in the screen space and sent to the rasterizer Rasterizer produces zero or more fragments, one fragment for each pixel covered by the triangle.

Looking for the full-text?

One or more pixel shaders, or in this example fragment processorcalculates the color or shade of gray for each fragment, typically using values from the texture memory Dynamic Reconstruction and Rendering (DRR) is a fast and flexible tomosynthesis image reconstruction and display implementation.

By leveraging the computational efficiency gains afforded by off. Dynamic reconstruction and rendering of 3D tomosynthesis images By leveraging the computational efficiency gains afforded by off-the-shelf GPU hardware, tomosynthesis reconstruction can be performed on demand at real-time, user-interactive frame rates.

A method of dynamically reconstructing three dimensional (3D) tomographic images from a set of projection images is disclosed. The method includes the steps of loading a set of projection images into a memory device, determining a reconstruction method for the set of projection images, reconstructing a 3D tomographic image from the set of projection images to be displayed to a user; and.

Dynamic Reconstruction and Rendering (DRR) is a fast and flexible tomosynthesis image reconstruction and display implementation. By leveraging the . Dynamic Reconstruction and Rendering (DRR) is a fast and flexible tomosynthesis image reconstruction and display implementation.

Dynamic reconstruction and rendering of 3d tomosynthesis images

By leveraging the computational efficiency gains afforded by off-the-shelf GPU hardware, tomosynthesis reconstruction can be performed on demand at real-time, user-interactive frame rates. Dynamic Reconstruction and Rendering of 3D Tomosynthesis Images Dynamic Reconstruction and Rendering (DRR) is a fast and flexible tomosynthesis image reconstruction and display implementation.

By leveraging the computational efficiency gains afforded by off-the-shelf GPU hardware.