PMOD Alzheimer's Discrimination Tool Introduction

<< Click to Display Table of Contents >>

Navigation:  »No topics above this level«

PMOD Alzheimer's Discrimination Tool Introduction

The PMOD Alzheimer's Discrimination Tool (PALZ) supports the fully automatic analysis of FDG brain PET scans acquired from subjects with clinical symptoms of Alzheimer's Dementia (AD). It is based on the data of a large multi-center trial and implements the proven discrimination methodology developed by Prof. K. Herholz et al. [1].


It is a prerequisite for the applicability of the analysis that the test subject suffers from clinical symptoms of AD. Assuming that this is the case, the PALZ tool performs the following processing steps with a loaded FDG brain PET scan:

The image is first stereotactically normalized using the methodology and the anatomical PET template of SPM99 [2].

The resulting image is then smoothed by a 12 mm Gaussian filter to minimize side-effects of varying scanner resolutions.

The image voxel values are divided by the voxel mean value, averaged within a mask representing voxels with AD-preserved activity. This operation results in normalized voxel values which are comparable across subjects.

In each image voxel the activity expected at the subject's age is calculated using voxel-wise age regression parameters.

The difference between the measured and the expected voxel values are transformed into t-values, resulting in a t-map [3].

The t-values of all abnormal voxels within an predefined AD-mask are summed, yielding a criterion related to AD-severity called AD t-sum. The AD mask represents all voxels that had shown a close correlation (at p<0.01 uncorrected) with the Mini Mental Status Examination (MMSE) in subjects with probable AD (see [1], Fig. 1). It includes major parts of the temporal and parietal association cortices (including the precuneus and posterior cingulate), as well as some lateral prefrontal areas.

The AD t-sum is compared with the 95% prediction limit (11089) established in the NEST-DD multi-center trial, and tested for significance of abnormality. The result of this test is the main outcome of the present AD discrimination.

The PET Score is calculated from the AD t-sum by reference to the upper normal limit and log transformation to approach a normal distribution: PET Score = log2((AD t-sum/11089 +1) [4].

For interactive reviewing, the t-maps can be displayed and fused with the normalized subject images. They can also be saved for use in an external analysis.

Additionally, a cluster analysis is performed for grouping of voxels with significantly reduced uptake [1]. Clusters of at least 216 contiguous voxels with p<0.05 (not corrected for multiple comparisons) are determined, ordered by their size in a table, and also made available as images.

For a thorough understanding of the discrimination analysis please refer to the NeuroImage article which exactly describes how the method was derived and validated [1].


AD t-sum [1]: The discrimination analysis described above provides the basis for a statistical test with null hypothesis "The AD t-sum is normal". As the distribution of the AD t-sum has been assessed in the control group, a 95% prediction limit and error probabilities were calculated. In the evaluation, the AD t-sum was shown to be a highly sensitive indicator of scan abnormality [1]. Therefore, if the AD t-sum is within the normal range it is unlikely that a tested subject has AD. Otherwise, if the AD t-sum is outside the 95% prediction limit, the null hypothesis is rejected ("AD t-sum within AD regions is abnormal"), and the error probability is stated (eg. "Error probability < 0.0001"). Such a finding, always in conjunction with clinical symptoms, supports a classification as AD.

PET Score [4]: The PET Score was shown to be a valid imaging biomarker for monitoring the progression of mild cognitive impairment (MCI) to Alzheimer's disease (AD). Its excellent test-retest reliability and signal strength is expected to allow substantially reducing the number of subjects or shortening of study duration in clinical trials.

Clusters: The outcome of the cluster analysis is not quantitative and solely intended for the visualization of concentrated area of abnormal voxels. Note that even in perfectly normal subjects up to 5% of all voxels may appear abnormal, and the appearance of a few abnormal clusters does not indicate any abnormality.


In 2009 the performance of the PALZ tool was assessed with a large number of cases from two independent databases (ADNI and NEST-DD) [5]. The databases differ regarding the demographics and the acquisition protocols as summarized below. The data sets were processed with PALZ, and the AD t-sum outcome used to calculate sensitivity and specificity.


Mild AD-subjects




102 controls,
age 76±4.9,
MMSE 28.9±1.1

89 subjects,
age 75.7±7.6
MMSE 23.5±2.1

Eyes open,
quiet darkened room, 3D

Sensitivity: 83%
Specificity: 78%


36 controls,
age 62±45;
MMSE 29.6±0.7

237 subjects,
age 70.8±8,
MMSE 20.9±4.4

Eyes closed

Sensitivity: 78%
Specificity: 94%

The main outcome of the study is the confirmation of the high accuracy of FDG PET with PALZ for the discrimination between AD subjects and normal controls. Furthermore it was demonstrated that although the eyes open condition results in higher occipital glucose consumption, there is no relevant effect on the AD t-sum. Therefore, PALZ can be equally applied to studies with eyes open or closed during the uptake phase. Only 7 out of 464 images had to be excluded from the analysis due to a failure of the normalization procedure.


The PALZ tool may only be used to analyze FDG brain scans of subjects with suspected AD. Note that the following guidelines must be observed to avoid invalid results:

A subject must not be younger than 49 years. For such subjects, the linear age regression might result in a (most probably too) high expected uptake in the frontal cortex. The consequence of such a situation would be a false positive finding.

PET images must be reconstructed with attenuation and scatter correction.

The stereotactic normalization may fail. In that case the subsequent discrimination calculations will be invalid. Therefore, the user is required to verify the outcome of stereotactic normalization.
Failure of normalization has been observed in the presence of severe brain atrophy. There are two indications of such a situation:

1.A shell-like distribution of the abnormal voxels along the boundary cortex/external liquor space.

2.Concentration of the abnormal voxels in the cingulate gyrus and the basal ganglia as a consequence of the severe widening of the inner liquor space.


The PALZ tool is not a general brain FDG analysis tool and thus not suited to search for non AD-related defects in FDG brain scans. Any other disease that also affects the association brain areas, which are abnormal in AD, may also lead to a significantly abnormal result. Scientific evidence is not yet sufficient for the interpretation of an abnormal AD t-sum in subjects without clinical symptoms of AD.


DISCLAIMER: PMOD is a software FOR RESEARCH USE ONLY (RUO) and must not be used for diagnosis or treatment of patients.