Abstract
1637
Objectives: To evaluate the use of functional volumes for partial volume correction (PVC) in PET imaging. For this purpose functional volumes determined on PET images using an adaptive Bayesian segmentation algorithm were used in combination with two PVC approaches (the Mutual Multiresolution Analysis (MMA) and Rousset).
Methods: The segmented functional volumes are obtained using the FLAB (Fuzzy Locally Adaptive Bayesian) algorithm, which has been previously shown to accurately segment functional volumes under variable noise and contrast conditions. In order to assess the proposed methodology a series of test images were employed. These included images of the IEC phantom (lesions sizes between 37mm and 10mm, contrast 8:1) acquired using a GEMINI PET/CT system and 15 simulated images of lesions derived based on real patient tumours, in order to have the ground truth in each case. For all of these images the FLAB algorithm was used to derive the functional volumes that were subsequently used in combination with the two partial volume correction algorithms. In the case of the MMA algorithm, which as a voxelwise PVC approach produces corrected images, region of interest analysis was performed to assess the improvement in quantitative accuracy.
Results: Significant improvement in quantitative accuracy (and associated image enhancement in the case of the MMA algorithm) was observed in our set of simulated tumours. In particular, the calculated activity in corrected lesions was 93.7%+/-5.9% of the ground truth activity and the local contrast at the boundaries was multiplied by a factor 2.3.
Conclusions: The use of functional volumes for partial volume correction allows improved quantitative accuracy leading at the same time, in the case of voxelwise based PVC algorithms, to images with enhanced contrast.
- Society of Nuclear Medicine, Inc.