Christian Rubbert MD, Abstract Co-Author: Fellowship funded, Koninklijke Philips NV

Jose Luis Vercher-Conejero MD, Abstract Co-Author: Fellowship funded, Koninklijke Philips NV

Nghi Co Nguyen MD, PhD, Abstract Co-Author: Research Grant, Koninklijke Philips NV

Fabian Wenzel, Abstract Co-Author: Employee, Koninklijke Philips NV

Oliver Steinbach PhD, Presenter: Employee, Koninklijke Philips NV

James K. O'Donnell MD, Abstract Co-Author: Research support, Koninklijke Philips NV Research support, Eli Lilly and Company Speakers Bureau, Astellas Group Speakers Bureau, Bayer AG Advisory Board, Eli Lilly and Company Advisory Board, Navidea Biopharmaceuticals, Inc

Early and definite diagnosis of Alzheimer’s disease (AD) is critical, as current treatment options under consideration are not free of safety concerns. Currently, a combination of clinical, neurological and neuro-psychological testing and imaging is used in diagnosis. Visual evaluation of FDG PET brain images is challenging. Hence, the Society of Nuclear Medicine (SNM) recommends augmentation by (semi-)automatic quantification approaches. This study evaluated the impact on diagnosis of AD in FDG-PET/CT when applying voxel-based statistical testing to 3D volumes, which have been stereotactially normalized using b-splines.

N = 94 subjects (50% AD and 50% normal) were selected from the Alzheimer’s disease Neuroimaging Initiative (ADNI) database. Two readers with 1 and 6 years of clinical experience classified FDG PET images, first by visual assessment of original images, then by rating stereotactically normalized 3D volumes, on which statistically significant areas of hypo-metabolism (Z-Scores, p<0.001) were displayed as an overlay. In a blinded reading, likelihood of diagnosis was split between normal, AD, FTD, LBD and other, adding to a total of 100%. Diagnostic categories were not limited to AD or Normal. 6 subjects were used for review and training. Sensitivity and specificity were calculated for each reader using confusion matrices.

Sensitivity and specificity for the most inexperienced reader increased using voxel-wise statistical testing as overlay: For normal subjects, sensitivity and specificity were 95% and 56% without and 98% and 61% with overlays. In AD subjects, sensitivity and specificity were 54% and 95%, which changed to 59% and 98% when using overlays. Accuracy increased from 75% to 78%. The more experienced reader showed a better specificity by 2% (59% vs. 61%) for normal subjects when using overlays.

Voxel-wise statistical testing may help especially inexperienced readers in the differential diagnosis of dementia. As opposed to previously published studies, this not only applies to the assessment of cortical surface projections, but also to the assessment of stereotactically normalized 3D volumes.

Differential diagnosis in dementia and especially Alzheimer’s disease is challenging and may be augmented by software.

Rubbert, C, Vercher-Conejero, J, Nguyen, N, Wenzel, F, Steinbach, O, O'Donnell, J, Voxel-wise Statistical Testing of FDG-PET/CT: Impact on differential Diagnosis of Dementia.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14045608.html