A support vector machine-based approach to guide the selection of a pseudo-reference region for brain PET...

Area of Science:

• Neuroimaging
• Machine Learning
• Radiochemistry

Background:

• Positron Emission Tomography (PET) imaging is crucial for studying brain disorders.
• Interscan and intersubject variability in PET data can complicate analysis.
• Identifying reliable pseudo-reference regions is essential for quantitative PET analysis.

Purpose of the Study:

• To develop and validate a Support Vector Machine (SVM) based method for identifying pseudo-reference regions in brain PET scans.
• To reduce interscan and intersubject variability in PET imaging.
• To evaluate the SVM approach across different PET tracers and patient cohorts.

Main Methods:

• A binary linear SVM classifier was trained using PET datasets from distinct subject groups.
• Pseudo-reference regions were identified based on regional contribution to the SVM classification score.
• The method was tested on three cohorts using ¹¹C-PiB, ¹¹C-UCB-J, and ¹⁸F-DPA-714 PET tracers.

Main Results:

• The SVM approach consistently identified cerebellum, brainstem, and subcortical white matter as pseudo-reference regions across different tracers and cohorts.
• Robust performance was observed even with a reduced number of subjects.
• The identified regions align with known low-signal areas in the brain.

Conclusions:

• The SVM-based method effectively identifies pseudo-reference regions for brain PET scans without prior assumptions.
• This approach offers a robust and data-driven solution for reducing variability in quantitative PET analysis.
• The method is applicable across various PET tracers and study designs, enhancing neuroimaging research.