An evaluation of Z-transform algorithms for identifying subject-specific abnormalities in neuroimaging data

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Psychiatric Disorders

Background:

• Subject-specific abnormalities (SSA) detection is vital for neuropsychiatric disorders.
• Understanding preprocessing effects on SSA identification is limited.
• Z-transform algorithms are commonly used for SSA analysis.

Purpose of the Study:

• Evaluate z-transform algorithms for SSA detection.
• Assess impact of data properties and preprocessing on SSA.
• Compare algorithm performance using simulated and DTI data.

Main Methods:

• Simulated data and diffusion tensor imaging (DTI) from 50 healthy controls.
• Evaluation of LOO, IDS, EZ-MAP, DisCo-Z, and ROB-Z algorithms.
• Analysis of distributional properties, sample size, and preprocessing steps.

Main Results:

• LOO, IDS, EZ-MAP, and DisCo-Z removed spurious differences; ROB-Z did not.
• LOO and IDS overestimated SSA across sample sizes and distributions.
• EZ-MAP and DisCo-Z showed better accuracy, except for skewed distributions.
• DTI analysis revealed preprocessing impacts on extrema counts.

Conclusions:

• EZ-MAP and DisCo-Z are more reliable for SSA detection than LOO and IDS.
• Preprocessing steps like blurring and registration influence SSA detection.
• Need for statistical comparison of SSA in controls or confidence intervals for patients.