FDG-PET contributes to the diagnosis and management of neurological diseases. In some of these diseases, pathological gray matter (GM) areas may have a reduced FDG uptake. Detection of these regions can be difficult and some remain undiscovered using visual assessment. The main reason for this detection problem is the relatively small thickness of GM compared to the spatial resolution of PET, known as the partial volume effect. We have developed an anatomy-based maximum-a-posteriori reconstruction algorithm (A-MAP) which corrects for this effect during the reconstruction using segmented magnetic resonance (MR) data. Monte-Carlo based 3-D brain software phantom simulations were used to investigate the influence of the strength of anatomy-based smoothing in GM, the influence of misaligned MR data, and the effect of local segmentation errors. A human observer study was designed to assess the detection performance of A-MAP versus post-smoothed maximum-likelihood (ML) reconstruction. We demonstrated the applicability of A-MAP using real patient data. The results for A-MAP showed improved recovery values and robustness for local segmentation errors. Misaligned MR data reduced the recovery values towards those obtained by post-smoothed ML, for small registration errors. In the human observer study, detection accuracy of hypometabolic regions was significantly improved using A-MAP, compared to post-smoothed ML (P < 0.004). The patient study confirmed the applicability of A-MAP in clinical practice.
Conclusion: A-MAP is a promising technique for voxel-based partial volume correction of FDG-PET of the human brain.