Template comparison for spatial normalization of [18F]flutemetamol brain PET images

Objectives [18F]flutemetamol (FLUT) is a PET tracer for in vivo quantification of brain β-amyloid. This work compares spatial normalization using different templates.

Methods 32 FLUT data sets from 13 healthy, 9 MCI and 10 AD subjects were normalized to MNI space by rigid intra-subject PET to MR MPRAGE registration, then by affine MR to T1 template registration. The combined rigid and affine transformations using FLUT and MR images formed a reference and compared with direct affine normalization of FLUT data sets using five different templates: a general FLUT template (FLUT build with all datasets), FLUT negative template (FLUT neg build with 17 datasets classified by an expert nuclear physician), FLUT positive template (FLUT pos build with 9 datasets expert classified) and a T1 and FDG template, both available in SPM. ‘Leave one out’ principle was used for the FLUT templates. Transformations were compared by calculating the Frobenius norm (FRO) after multiplication with the inverse reference transformation (FRO identity matrix = 2).

Results The general FLUT template showed the lowest (best) FRO (3.42±0.22), significantly lower than for the FDG template (18.81±3.19) and T1 template (4.06±0.24). Within group results for FLUT negative data sets (avg FRO ± SD FDG:30.31±3.26, T1:3.91±0.33, FLUT:3.44±0.31, FLUTneg:3.04±0.18, FLUTpos:6.61±1.66) showed no significant difference for other templates when compared with the general FLUT template, except for FDG template. Within group results for FLUT positive data sets (avg FRO ± SD for FDG:4.36±4.49, T1:4.12±0.46, FLUT:3.18±0.43, FLUTneg:4.57±0.25, FLUTpos:3.00±2.28) showed significant difference for the T1 template and FLUT negative template when compared with the general FLUT template.

Conclusions A general FLUT template provides accurate affine normalizations for all FLUT datasets and can be accurately used with positive and negative datasets. The SPM T1 template provides good normalization results while the SPM FDG template is not generally suitable