Abstract
573
Objectives: PET/CT misalignment is widely observed in patient studies and can introduce image artifacts and quantitation bias that adversely impact diagnosis. Automatic detection and correction of PET/CT misalignment is critical but remains as a challenging goal for clinical applications. In previous work, we have shown that PET/CT misalignment leads to count redistribution in CT-based attenuation correction (CTAC) and hence image artifacts and quantitative bias. TOF provides the information of event origination and constraints the count redistribution. Therefore, incorporating the TOF information in image reconstruction can systematically reduce image artifacts and quantitative bias. Using this unique advantage of TOF, we propose a technique to automatically detect and correct for PET/CT misalignment.
Methods: For misalignment detection, the data with high-resolution TOF is reconstructed twice with CTAC. The first reconstruction (R1) models the high-resolution TOF of the data to obtain a high TOF resolution image. The second reconstruction (R2) does not model TOF to obtain a non-TOF image. Then the square of the difference image of R1 and R2, i.e., (R1-R2)2, is calculated. PET/CT misalignment in the region where soft tissue in CT overlays with lung tissue in PET will show as a strong bright band, the width of which indicates the magnitude of the misalignment in general. For correction, the data is reconstructed a third time (R3), assuming the TOF resolution is lower than the true TOF resolution of the data. R3 reconstruction (1) intentionally degrades the TOF resolution of the data using a Gaussian convolution kernel and (2) models the degraded TOF resolution in the iterative reconstruction. The TOF kernel of the original data, when convolved with the Gaussian kernel, gives the TOF kernel of the degraded TOF. R3 is hence an effectively low TOF resolution image. Misalignment correction can be done using image algebra of R1, R2, and R3 in image domain. The TOF resolution for R3 reconstruction and the image algebra can both be optimized based on the system TOF and related specifications. For demonstration purpose in this work, we simply used R1 + (R3-R2) to approximate the corrected image. Simulation studies with lesions in lungs were used with TOF resolution of 320 ps and R3 reconstruction used TOF of 640 ps. Patient data was from a digital PET/CT system with 325 ps TOF resolution and artificial PET/CT misalignment was introduced.
Results: In both simulation and patient studies, the proposed technique automatically detected PET/CT misalignment in the torso region. In the simulation studies, the automatic correction (a) effectively removed the strong band artifacts where soft tissue in CT was overlaid with lung tissues in PET, (b) effectively removed ghost lesions, and (c) significantly improved the lesion quantitation. Simulation also showed that the automatic correction would not be effective if the original TOF resolution was poor (e.g., ~600 ps) due to the small improvement over non-TOF.
Conclusion: The described technique used high-resolution TOF information with advanced reconstruction schemes for automatic PET/CT misalignment detection and correction. The automatic correction effectively removed strong band artifacts at soft tissue/lung boundaries and significantly improved lesion quantitation. Results also suggest that high TOF resolution is required for the automatic correction to be effective. Research Support: This work is supported by Philips.