Revisiting intensity normalization in brain [18F]FDG PET: What is reliable in autoimmune encephalitis?
Purpose Data-driven intensity normalization has emerged as an alternative to proportional scaling (PS) and reference-region methods in brain [18F]FDG PET. However, no consensus exists for autoimmune encephalitis (AE), whose variable metabolic patterns and lack of a reliable disease-free reference region complicate normalization. We compared three methods; PS, iterative PS (iPS), and pons-based reference-region (RR) normalization, in patients with definite AE (n = 29, 42 ± 20 years) and healthy controls (HC; n = 53, 46 ± 15 y/o) examined with brain [18F]FDG PET at diagnosis. Methods Three sets of normalized [18F]FDG PET images (2 MBq/kg, Biograph mCT Flow PET/CT system, Siemens Healthcare) were generated using: (1) PS with an eroded AAL grey-matter mask; (2) iPS, applying PS then excluding voxels deviating from controls (SPM12, F-contrast p < 0.01 uncorrected) to create a subject-specific mask; and (3) RR, normalizing to mean pons uptake. Effects were assessed through voxel-based AE vs. HC comparisons in SPM12. Two limbic AE cases were analyzed longitudinally. Results PS and iPS revealed basal ganglia and mesiotemporal hypermetabolism in AE vs. HC that RR failed to detect (p < 0.05 FWE corrected). All methods identified cortical hypometabolism, slightly more extensive with RR. PS and iPS yielded broadly similar results. Follow-up showed that iPS and PS better captured initial abnormalities than RR. Conclusion PS and iterative PS may provide more robust normalization than RR in AE. As interest in neuroinflammatory disorders grows, standardized normalization protocols are needed to ensure consistency across studies.