Robust fitting of [11C]-WAY-100635 PET data

Area of Science:

• Neuroscience
• Medical Imaging
• Biostatistics

Background:

• Positron emission tomography (PET) data analysis commonly uses least squares (LS) fitting.
• LS is sensitive to outliers, potentially compromising parameter estimate accuracy.
• Quantile regression (QR) offers robustness against outliers in statistical modeling.

Purpose of the Study:

• To investigate if quantile regression (QR) improves parameter estimate accuracy in PET modeling compared to least squares (LS).
• To assess if QR reduces intersubject variance in distribution volume (V(T)) estimates.
• To evaluate the impact of QR on statistical power in group analyses.

Main Methods:

• Applied quantile regression (QR) and least squares (LS) to PET time-activity curve data.
• Analyzed data from 49 healthy controls using [(11)C]-WAY-100635.
• Compared the intersubject variance of distribution volume (V(T)) estimates between QR and LS methods.

Main Results:

• Quantile regression (QR) demonstrated a reduction in the standard deviation of V(T) estimates (0.08% to 3.24%) compared to LS.
• Average V(T) values remained largely unchanged between QR and LS methods.
• QR's variance reduction enhances statistical power for group analyses, potentially reducing subject requirements.

Conclusions:

• Quantile regression (QR) provides more accurate and robust parameter estimates in PET modeling than traditional least squares (LS).
• QR's reduced intersubject variance improves statistical efficiency, enabling more powerful group comparisons with fewer subjects.
• This advanced PET analysis technique offers significant advantages for neuroimaging research without necessitating additional hardware.