Quantitative Intensity Harmonization of Dopamine Transporter SPECT Images Using Gamma Mixture Models

Area of Science:

• Neuroimaging
• Radiochemistry
• Medical Physics

Background:

• Dopamine transporter (DAT) single-photon emission computed tomography (SPECT) imaging is crucial for neurological studies.
• Variations in imaging site, device, and settings lead to significant intensity profile heterogeneity in DAT SPECT images.
• Current analysis methods, like the striatal binding ratio (SBR), do not adequately address this heterogeneity, impacting cross-site interpretation.

Purpose of the Study:

• To develop and validate a voxel-based automated approach for intensity normalization of DAT SPECT images.
• To improve the cross-session and cross-site interpretability of DAT SPECT data.
• To enhance the reliability of quantitative analysis and downstream machine learning applications.

Main Methods:

• A novel normalization method using the cumulative density function (CDF) of a Gamma distribution (GDCF) was applied to reparametrize voxel intensities.
• The method was tested on 1342 DAT SPECT images from the PPMI repository, encompassing 7 camera models and 24 sites.
• Quantification was compared across different cameras using raw intensities, SBR, and GDCF-normalized values. A classification task was used as a proof-of-concept.

Main Results:

• Significant differences in raw striatal intensities and SBR were observed across camera models.
• GDCF normalization effectively reduced these differences, constraining striatal quantification values to [0, 1].
• The automated GDCF method achieved high classification accuracy (AUC = 0.94-0.98) between controls and Parkinson's disease patients.

Conclusions:

• The GDCF normalization method provides an automated and effective way to standardize DAT SPECT image intensity.
• This standardization is crucial for developing unbiased algorithms that utilize multicenter datasets.
• GDCF normalization represents a key pre-processing step for robust DAT SPECT image analysis.