4D Median Root Prior in PET Image Reconstruction for Noise-Suppression in Dynamic Neuroimaging

Ziel/Aim Dynamic neuroimaging using PET suffers from noise which increases with increasing temporal resolution, i.e. lower count statistics per frame. Thus, noise suppression techniques are desirable. Maximum a posteriori (MAP) PET image reconstruction in combination with the Median Root Prior (MRP) is an option [1]. Now, we have extended the 3D MRP considering the spatial domain by additionally considering the time domain. This is motivated by the fact, that Time-Activity-Curves (TAC) are expected to be smooth (when excluding the early injection phase). Thus, the new 4D MRP uses temporal information to further suppress noise.

Methodik/Methods We extend the 3D MRP (addressing 3^3=27 neighbor voxels in the spatial domain) to the time domain by additionally considering the same voxels in the previous (f-1) and the next frame (f+1) with respect to each frame f. Then, the median covers 3 x 27 = 81 voxels. Instead of independent reconstruction of any frame, the 4D MRP requires all image iteration updates in parallel to coherently calculate the 4D MRP for each iteration.

We used brain tumor data with O-(2-[¹⁸F]fluoroethyl)-L-tyrosine (FET) which was acquired on the Siemens hybrid MR/BrainPET scanner [2]. 14 dynamic frames (3 minutes post-injection) each of 200 sec frame length have been reconstructed using PRESTO [3] with an isotropic voxel size of 1.25³ mm³. TACs are compared by using no MRP, conventional 3D MRP and 4D MRP.

Ergebnisse/Results FET brain tumor images and corresponding TACs indicate a superior noise suppression for the 4D MRP while resolution is not visibly degraded.

Schlussfolgerungen/Conclusions The exploitation of inter-frame smoothness results in a noticeable boost of noise suppression compared to conventional 3D MRP alone. For example, TACs provide information about tumor malignancy [4], thus diagnosis can have benefit when using 4D MRP.

• Literatur/References:

• 1 Eur J Nucl Med, 1997; , 24 , No (03) , 258-265.
• 3 Ieee Tmi, 2011; , 30 , 879-892.
• 2 Ieee Trpms, 2019; , 3 , No (03) , 352-363.
• 4 J Nucl Med Mol Img, 2007; , 34 , 2082-2087.