Hybrid segmentation-atlas method for PET-MRI attenuation correction

One of the major unsolved issues of PET-MRI is the PET attenuation correction using MR images. Conventionally, in PET or PET-CT, attenuation maps (μ-maps) have been obtained by a PET transmission scan, or a CT scan. For PET-MRI, in order to obtain μ-maps from MR images, many studies have been reported, and they can be classified into two methods; the atlas-based method (ABM) and the segmentation-based method (SBM). In the ABM, individual differences such as lesions are not supported. In the SBM, it is difficult to discriminate bone and air, which have large differences in their attenuation coefficients, because these tissues have similar MR signal values in the T1 weighted (T1w) MR images. In this work, therefore, we proposed a hybrid segmentation-atlas method (HSAM) to utilize the advantages and compensate for the disadvantages of both the ABM and the SBM. At first, the proposed method follows the SBM approach. In the bone and air regions where T1w MRI signals are similar and low, the HSAM uses information from a standard μ-map obtained through the ABM. For evaluation, the head data from 6 healthy volunteers were obtained by PET (ECAT Exact HR+) and MRI (Philips Intera 1.5T). We estimated μ-maps by the ABM, the SBM and the HSAM, and PET images were reconstructed though attenuation correction with those μ-maps. In comparison of the μ-maps, the HSAM and ABM outperformed the SBM. In comparison of the final PET images, a similar tendency was seen. For patient data, which would contain different distribution from the database, the HSAM is expected to outperform the ABM.