Correcting power and p-value calculations for bias in diffusion tensor imaging.

Area of Science:

• Neuroimaging
• Biostatistics
• Medical Physics

Background:

• Diffusion Tensor Imaging (DTI) generates quantitative maps of tissue microarchitecture, such as fractional anisotropy (FA).
• DTI data processing involves computational methods susceptible to random distortions, including variance and bias.
• Conventional statistical methods often overlook bias, focusing primarily on variability.

Purpose of the Study:

• To quantitatively assess the impact of bias in DTI on hypothesis testing properties, specifically statistical power and alpha rates.
• To evaluate the effectiveness of the SIMEX (Simulation and Extrapolation) technique for estimating bias in DTI.
• To analyze the effects of bias on spatially varying power and alpha rates in an empirical DTI study.

Main Methods:

• Theoretical evaluation of bias effects on hypothesis testing.
• Simulation studies using the SIMEX technique to estimate DTI bias.
• Empirical analysis of bias impacts on power and alpha rates in a 21-subject DTI dataset.

Main Results:

• Bias significantly inflates alpha rates (Type I errors).
• Bias distorts the power curve, leading to substantial power loss.
• Adverse effects of bias persist even when group bias differences are minimal.
• Bias impacts are evident in spatially varying power and alpha rates.

Conclusions:

• Bias in DTI negatively affects hypothesis testing, compromising the reliability of study findings.
• Bias estimation and correction are essential for accurate statistical inference in DTI research.
• Properly accounting for bias in power and p-value calculations can mitigate adverse effects.