Attenuation of gamma photons in positron emission tomography (PET) results in a loss of signal that adversely affects the quantitation of PET images. To perform attenuation correction, the distribution of tissue electron density-dependent linear attenuation coefficient (LAC) values is required¹. Most methods that provide continuous-valued LAC maps utilize atlas registrations and complex algorithms. This approach is time-consuming and may not be suitable for patients whose anatomy cannot be represented well by population data. Alternately, the current vendor-provided UTE method employs a dual-echo UTE acquisition for tissue segmentation followed by an assignment of constant LAC values for soft tissue, bone, and air. However, inaccuracy in tissue segmentation and representation of a wide range of tissues using only a few discrete LACs leads to PET quantification errors. In this study, we aim to address both issues using a quantitative T₁ mapping method derived from a dual flip angle and dual echo UTE sequence.Acquisition. CT and PET/MR images were acquired from volunteers (n=23) who provided informed, written consent. CT images were acquired at 120 kVp and spatial resolution=0.59x0.59x3 mm³. PET images were acquired with ¹⁸F-Florbetapir tracer and reconstructed to spatial resolution=2.09x2.09x2.03 mm³. Dual flip angle (θ₁=3°, θ₂=25°) and dual echo (TE₁=0.07 ms, TE₂=3.69 ms) UTE-MR images were acquired with TR=9 ms, radial spokes=13,000, spatial resolution=1.56x1.56x1.56 mm³ and total acquisition time=3:54 min. The echo times were chosen with water and fat in-phase (TE₁) and out-of-phase (TE₂), and the utilized flip angles were chosen using optimized Ernst angles to maximize imaging contrast among bone, soft tissue, CSF and air. Processing. R₁ (1/T₁) maps were computed using the dual flip angle UTE images acquired at TE₁ (UTE₁). Regions of air were identified using joint thresholding of UTE₁ images acquired with θ=3° and θ=25°. Regions of bone, GM, WM, and CSF were segmented using patient-specific thresholding of R₁ maps. Fat and water are distinguished using a two-point Dixon approach applied to UTE images acquired at TE₁ and at TE₂ (UTE₂) with θ=3°. Linear relationships between R1 and CT Hounsfield units were derived for each of GM, WM, and CSF, while a logarithmic relationship was derived for bone; these relationships were then used to convert R₁ to continuous-valued PET attenuation coefficients. Constant LAC values are assigned to regions of fat and air. The method was dubbed TESLA for T1-enhanced segmentation and selection of linear attenuation coefficients. Validation. TESLA attenuation maps were generated for each subject using a leave-one-out approach. PET images were reconstructed using the vendor-provided software (e7Tools, Siemens) with four attenuation maps: (1) gold standard CT-based map, (2) vendor UTE map (vUTE), (3) TESLA map with constant LAC values for each tissue (TESLAseg), and (4) proposed TESLA map (Figure1).Mean (±standard deviation) whole-brain errors of PET reconstructions computed against the gold standard were 7.76% (±1.77) for the vUTE method, 4.65% (±1.67) for the TESLA_seg method, and 3.02% (±1.13) for the TESLA method. The mean errors were significantly (p<0.01) lower for the TESLA method than the other two methods. Percent-error maps from a representative subject (Figure 2) illustrate that the proposed TESLA method achieves favorable error distributions in most brain regions whereas the vUTE method largely underestimates the PET signal in many brain regions.The proposed TESLA method utilizes images from a dual flip angle, dual echo UTE-MR acquisition to segment air, bone, GM, WM, CSF, fat and soft tissue. A conversion from MR relaxation rate R₁ is then utilized to derive continuous-valued LACs to major tissues in the head/neck. The computation of R₁ allows for the separation of and assignment of continuous-valued LACs to brain tissues in a manner which is not possible using relaxation rate R₂^*, which has previously been used to provide continuous-valued LACs for bone only^2-4. While the LAC differences between GM, WM, and CSF are relatively low, the large number of brain voxels may have a cumulative effect on PET image accuracy. The method has been validated in PET data from 23 subjects and has been shown to outperform the vendor UTE method in PET reconstruction accuracy. The use of a second flip angle increases the total acquisition time relative to the vUTE method, but this increase was mitigated by reductions in TR and the number of radial spokes.The proposed TESLA method provides PET attenuation maps with continuous-valued LACs for bone, GM, WM, and CSF using only patient MR images and conversion equations derived a priori. It outperforms the vendor UTE method in whole-brain PET reconstruction accuracy and has been fully automated to enable incorporation into the PET/MRI clinical workflow.