PET/MRI is a hybrid imaging modality that combines PET and MRI in a single integrated platform. This allows for simultaneous acquisition of PET and MRI data. A major challenge of PET/MRI is the generation of an accurate attenuation coefficient (AC) map. Furthermore, the skeleton is one of the most common organs for advanced cancer to metastasize[4]. In order to properly stage or assess treatment response, lesion uptake quantification must be as accurate as possible.

The current clinical standard for MR-based attenuation correction (MRAC) in an extracranial region is a fat-water map generated using a Dixon-type sequence[1]. This sequence allows fast fat-water separation, however the generated attenuation map is unable to differentiate between bone and intrabody air, which is critical for assessing bone lesions. In contrast, ultrashort and zero echo time (UTE and ZTE) sequences are able to distinguish bone from air[5], hence, we developed a hybrid MRAC technique that involves bone segmentation from a ZTE MRI, and water and fat from the Dixon MRI. In this work, we specifically investigated the effect of bone attenuation on pelvic lesion uptake.

Five patients underwent simultaneous ¹⁸F-FDG PET and MRI using a 3-tesla time-of-flight PET/MRI system (GE Signa PET/MR).

MR images were acquired using a ZTE pulse sequence (dead time = 8 μs, FOV = 34 cm × 34cm, resolution = 2 mm × 2 mm × 2 mm isotropic, BW = ±62.5 kHz, FA = 1°, PD-weighted, TR = 527 μs, spokes/seg = 256), and a two-point Dixon pulse sequence (FOV = 500 mm, resolution =1.95 mm × 1.95 mm, slice thickness = 5.2 mm, TE = [1.15 ms, 2.3 ms], TR = 4.05 ms).

Bone was segmented from ZTE images as follows: N4 bias correction was applied to the images to eliminate intensity inhomogeneities[6]. The bone signal was enhanced and an initial bone segmentation map was generated using global thresholding. The segmentation map was then manually corrected based on the expected pelvic bone anatomy to remove residual air, similar to how an atlas-based segmentation method would perform. The resulting bone map was assigned a single HU value (1000 HU) and combined with the Dixon pseudo-CT generated by the offline reconstruction toolbox. The pseudo-CTs are converted to an MRAC map using a bilinear model.

PET image reconstruction was performed using a time-of-flight ordered subsets expectation maximization (TOF-OSEM) algorithm (FOV = 600 mm, 2 iterations, 28 subsets, matrix size = 192 × 192, slice thickness = 2.78 mm) with the hybrid MRAC map and Dixon-only MRAC map.

For analysis, lesions were separated into two classes: bone lesions, which were surrounded by bone, and soft-tissue lesions, at least 10 mm away from the nearest bony structure. Representative images for each lesion type are shown in Figure 1.

The maximum standardized uptake values (SUV_max) were measured for quantitative analysis in the lesion volumes segmented by a region-growing algorithm.

A 3D volume rendering of the bone segmented from the ZTE images is shown in Figure 2. The rendered bone information is able to accurately depict the bone in the pelvis. The bone information is combined with the Dixon MRAC map to produce the hybrid MRAC map (Figure 3).

PET SUV difference images are shown in Figure 4. The SUV_max for each lesion is listed in Table 1a, and the % underestimations are summarized in Table 1b. SUV_max in bone lesions have an underestimation of 10-20% by the Dixon MRAC when compared to the hybrid MRAC. Soft-tissue lesions were less affected, with the underestimation under 10%.

In our study, the underestimation of uptake for bone lesions without bone information was clinically significant (defined to be greater than 10%), while uptake in soft tissue lesions was not as clinically significant. The magnitude of the underestimation we observed is consistent with current literature[2,3], however these studies investigated the effects of bone based on simulating MRAC maps using PET/CT data, whereas our hybrid MRAC is derived from MR data.

Our study highlights the necessity of including bone information when quantifying uptake in bone metastases. The inclusion of bone information in MRAC is less important for soft tissue lesions. However, the impact of bone information on uptake quantification will depend on how close the lesion is to regions of high bone density.A hybrid MRAC map (with fat-water soft tissue mapping, bone, and air) was successfully generated from combined ZTE and Dixon acquisitions. The Dixon-only MRAC significantly underestimates uptake in bone lesions, indicating that bone cannot be ignored in MRAC when the uptake measurement in bone lesions is clinically important.