A tissue-fraction estimation-based segmentation method for quantitative dopamine transporter SPECT.

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Medical Image Analysis

Background:

• Accurate segmentation of caudate, putamen, and globus pallidus (GP) in DaT-SPECT images is vital for Parkinson's disease severity assessment.
• Partial-volume effects (PVEs) and finite tissue effects (TFEs) in SPECT imaging present significant challenges to precise region segmentation.
• There is a critical need for advanced methods to overcome PVEs and TFEs for accurate DaT-SPECT image analysis.

Purpose of the Study:

• To design and evaluate a fully automated segmentation method for caudate, putamen, and GP from DaT-SPECT images.
• The method is based on tissue-fraction estimation to address segmentation challenges.
• Objective evaluation of the automated segmentation's performance is a key goal.

Main Methods:

• A supervised deep-learning approach was employed to estimate the posterior mean of fractional volumes within each voxel.
• A cost function minimizing binary cross-entropy loss was used for fractional volume estimation.
• Training data for true fractional volumes were derived from existing clinical MRI datasets.

Main Results:

• The method achieved high accuracy in segmenting caudate, putamen, and GP, with mean Dice similarity coefficients around 0.80 in simulations.
• It significantly outperformed other segmentation methods evaluated.
• Reliable quantification of regional uptake was achieved, with normalized root mean square errors below 20%, and around 10% for caudate and putamen.

Conclusions:

• The developed tissue-fraction estimation method enables accurate segmentation of key brain regions in DaT-SPECT images.
• It also allows for reliable quantification of dopamine transporter uptake, essential for Parkinson's disease research.
• Further validation using physical phantoms and patient studies is recommended.