The state of the art in computed tomography in material science applications uses Filtered Back Projection (FBP) as the reconstruction algorithm and acquires views progressively from 0 to 180 degrees in fixed angular steps. In medical applications, by using Model Based Iterative Reconstructions (MBIR) instead of FBP, significant gains in reconstruction quality has been achieved. However, high temporal resolution reconstructions of time varying objects is still an unexplored field.
MBIR has a forward model which models the projections as a function of the attenuation coefficients and a prior model which models the self similarity between the neighboring voxels. Also, MBIR is more robust to noise because it uses a noise model.
Our research aims to achieve high temporal resolution reconstructions using MBIR based reconstruction algorithm and a interlaced view sampling methodology. The projection acquisition process is modeled as a time sequential process where every projection is acquired at a different point in time. Also, a new interlaced view sampling method is proposed in which several consecutive sub-frames of views are interlaced in angle. All the sub-frames combine to form a frame. By combining the new view sampling method with MBIR reconstruction, high temporal resolution reconstructions can be achieved.
However, simulation reveals that combining interlaced views with traditional reconstruction algorithms does not lead to any gains. In fact, with FBP, we observed a loss in quality. Thus, it is synergy between the MBIR algorithm and the interlaced views which results in significant gains in temporal resolution without loss of reconstruction quality.
The cover image gives a comparison between the new method and conventional FBP. It can be seen that MBIR with interlaced views provides very good temporal resolution.