Neutron Tomography

Tomography is a method which provides cross-sectional images of an object from transmission data, measured by irradiating it from many different directions. Tomographic imaging in a mathematical sense deals with reconstructing an image from its projections. Here the projection at a given angle represents the integral of the image in the direction specified by that angle.

The non-destructive analysis of an object by neutron radiography is mostly done taking one or more 2D parallel projections. In some cases however the transmission properties of the object seen from any angle are looked for. This can be achieved by rotating the object in small angular steps over 180^o and calculating tomographic slices using the inverse Radon transform. The excellent beam collimation properties of the NEUTRA beamline can be used to provide 3D transmission volume data from a series of 2D parallel projections of an object.

Components

The 150-300 times repeated turning of the object and acquiring a neutron transmission image is implemented in an automatic procedure connecting a camera with the rotating table and a neutron beam monitor. We use a nitrogen cooled CCD camera system (Astromed 3200 by PerkinElmer Life Sciences, Cambridge U.K.), which can be run in looping mode and allows for an external trigger signal. Depending on the object size (maximum weight/dimension 20kg/25cm), and the camera lens in use, it provides 512x512 pixel 16bit images with a resolution of 0.2 - 0.5 mm. Exposure times in the range from 5 - 60 seconds are imposed by the neutron flux level of SINQ and the objects neutron transmission given by its thickness and isotopic composition. Objects consisting mainly of strongly neutron absorbing isotopes and with a large length/thickness ratio are not suitable for neutron tomography.

A rotating table (Franke, Aalen, Germany, type 12062) is driven by a 2-phase stepping motor controlled by its own steering unit (in house PSI AEI assembly) and is supplemented by an additional goniometer. It allows a smooth and reproducible angular positioning. In order to avoid direct neutron illumination of the motor unit, the objects are fixed on a table supported by an aluminium extension pipe. The rotating table unit itself is mounted onto a 2-axis tilting table, which in turn sits on a x-y axis motion table used to position the sample precisely with regard to the CCD camera's field of view.

Data Acquisition

Due to the proton beam induced intensity variations of the spallation source, the neutron flux level at the sample position is monitored indirectly by measuring the g-background radiation originating from the sample and the beam dump. It is checked before and during the camera shutter is open, in order to guarantee sufficient neutron fluence for a single exposure i.e. a fluence threshold is set at about 80% nominal fluence, below which an image acquisition will be repeated. This rough measure suffices, because a single image is later on normalized with regard to an open beam region of interest of its own.

A LabView based software interface implemented on a PC allows to set up any rotation sequence. It sends the motion commands to the motor control unit, processes the neutron monitor signal and communicates with the CCD control software watching for the camera shutter trigger signal and providing the shutter close signal.

Data Processing and CT Reconstruction

There are image preprocessing steps needed before the projection data can be input to the reconstruction algorithm. The images have to be normalized for equal neutron exposure and corrected for spatial variations in the incident neutron beam and the scintillator screen (flatfield correction). The flatfield correction is performed dividing a sample projection image by an image without sample (flatfield image). The CCD chip of the camera is quite sensitive for gamma ray's, which show as bright spots randomly distributed in the projection images. These have to be removed by e.g. median filtering the flatfield and projection images. Occasionally images have to be centered after acquisition due to imperfect positioning of the turntable rotation axis. Finally the corrected ratio (object image / flatfield image ) for all angles is put into the so called sinogram volume dataset (size: nx X ny pixels X number of angles).

Individual transversal ( i.e. perpendicular to the rotation axis ) slices of the object can now be reconstructed using a filtered backprojection implementation of the inverse Radon transform (see e.g. : "Principles of Computerized Tomographic Imaging"). Finally all slices are collected in an image stack representing the 3D volume data of the neutron attenuation properties of the object, which can be visualized using 3D rendering software ( e.g. http://www.volumegraphics.com ). In this way parts of the object with different neutron attenuation characteristics can for example cut out (segmentation) or slices of arbitrary angle can be made visible inside the object.