Digital Detector Arrays (DDA)—also known as flat panel detectors—are increasingly becoming the detector of choice for industrial radiography (2D) and Cone Beam Computed Tomography (CBCT) (3D). Scattered radiation from the object being scanned reduces the achievable contrast resolution and can result in long scan times for detection of flaws assuming the object can be scanned at all. To overcome this challenge, Varex has extended its scatter correction model from the image processing pipeline of the Cone Beam Software Toolkit (CST)--originally developed for the medical market--to the higher energies, wider ranges of materials and imaging geometries of the industrial radiography market. Offering the latest software tools for 2D image correction and 3D image reconstruction, this toolkit includes a complete, customizable, flexible software library to help flat panel customers improve their radiographs and to implement Cone Beam CT (CBCT).
Image Processing Pipeline
The Varex pipeline includes a set of Windows DLL software libraries that perform image corrections to 2D radiographs and, if desired, CBCT image reconstruction. In addition, our image processing pipeline is easily accessible from our OEM’s .NET and C++ host applications and includes 32-bit and 64-bit versions with a stable, easy to use API. Most configurations are done via XML files, and the pipeline supports simultaneous acquisition and processing.
Most importantly, the pipeline offers flexibility. Customers can choose all Varex components (and select their own steps); a Varex pipeline with developer plug-ins; or a developer pipeline with Varex plug-ins. While the Varex pipeline includes improved beam hardening and lag corrections for industrial imaging, its enhanced scatter correction is customizable based on the material type and provides appropriate scatter kernels at different thickness steps through the object.
The Problem: Large Scatter-to-Primary Ratios
The scatter-to-primary ratios in digital detector array imaging are often high (>1) due to the large X-ray cone angle and detector entrance aperture. Having more scatter than primary signal leads to the washing out of details in the images, much like an overexposed photo. Large scatter-to-primary ratios also cause cupping (in addition to beam hardening) in the density profiles of uniform objects and streaking between high- and low-density objects.
The Correction: Kernel Method
Working with customers, Varex has developed a scatter kernel correction based on a customized model that takes into account the X-ray source energy as well as the imaging geometry to determine a scatter correction at the appropriate thicknesses in a material. Using a computationally efficient deconvolution method, scatter kernel corrections are based on pre-computed Monte Carlo simulations and can be fine-tuned with empirical measurements.
The scatter correction model includes three steps: 1. Make initial scatter estimate; 2. Subtract this initial scatter estimate from the detector signal to get the primary estimate; 3. Convolve the primary estimates with kernels to refine the scatter estimate. Steps 2 and 3 are iterative to achieve the final result.
The Kernel Method for scatter correction can be applied to 2D images reconstructed in 3D as well as 2D radiography. While we cannot completely eliminate scatter, we can mitigate its impact to greatly improve image quality.
Other Advances in Varex’s CST Software Suite
Over the past two years, Varex has integrated many more items into our software libraries that are appropriate for industrial imaging, from water up to Inconel 718 with multiple industrial energies and a rich array of spectra and attenuation coefficients ranging from 160 KeV up to 1 MeV.
In the future, we plan to include statistical reconstruction in our 2.0 CST Suite. An iterative technique, statistical reconstruction may slightly increase reconstruction time but can be optimized to provide better contrast and resolution performance simultaneously, compared to FDK methods where you choose one or the other. Other benefits include cone beam artifact reduction and fewer projections to achieve the same FDK quality, resulting in increased part throughput.