摘要
Introduction & Background
Brain PET on LAFOV PET/CT requires CT coverage of the full PET sensitivity range, but extending beyond the head adds unnecessary radiation. We evaluated whether deep-learning μ-map (DLμ) splicing enables full-axial attenuation correction from head-limited CT while maintaining quantitative accuracy across metabolic, dopaminergic, amyloid, and tau tracers.
Methods
Five head-centered acquisitions on a uMI Panorama GS LAFOV PET/CT: 18F-FDG, 18F-FDOPA, 18F-MK6240, and two 18F-florbetapir scans. CT covered 148 cm; a 20-cm head segment was extracted to simulate limited coverage. The DLμ model was trained on anatomy-derived synthetic PET-like images and applied to non-attenuation-corrected TOF images to predict μ-map. Reconstructions used (A) CT-derived attenuation map (CTμ) with PET events restricted to the corresponding segment; (B) full-axial PET events with zero-padded CTμ; (C) full-axial PET events with DLμ spliced into missing regions (proposed); and (D) full-axial PET events with full-coverage CTμ. Regional SUV bias relative to D was assessed in 11 AAL116 brain regions. Cerebellar coefficient of variation (CoV) was measured.
Results
Compared with A, method C retained more than 58 times as many total PET events, including events not traversing the brain, and reduced cerebellar CoV from 27.3% to 22.6%. Across 55 regional comparisons (11 regions × five acquisitions), C yielded the lowest median absolute SUV bias (0.93%; interquartile range [IQR], 0.28%–1.60%), compared with A (1.63%; IQR, 0.90%–3.86%) and B (8.93%; IQR, 5.19%–11.91%). Absolute SUV bias with C was within 2% in 90.9% and within 5% in all comparisons, versus 54.5% and 80.0% for A. Method C outperformed A in 39 of 55 comparisons and provided the closest agreement with reference in four of five acquisitions.
Conclusion
DLμ splicing retained full-coverage PET events and improved quantitative accuracy over restricted-event reconstruction under simulated head-limited CT. This supports exploitation of LAFOV sensitivity without extending clinical CT coverage beyond the head, potentially reducing unnecessary radiation.
