Comparative reliability analysis of publicly available software packages for automatic intracranial volume estimation

Area of Science:

• Neuroimaging
• Brain Volume Analysis
• Medical Software Evaluation

Background:

• Intracranial volume (ICV) is crucial for brain research, often used as a covariate in inter-subject studies.
• Accurate ICV estimation is vital for reliable neuroimaging analysis.
• Automated software offers efficiency but requires validation across diverse populations.

Purpose of the Study:

• To evaluate the performance of four automated software packages (AFNI, Freesurfer, FSL, SPM) for estimating intracranial volume (ICV).
• To determine if software performance varies across different subject groups: adult controls, adults with dementia, pediatric controls, and pediatric epilepsy patients.
• To assess the impact of different MRI scanner field strengths (1.5T and 3T) on ICV estimation accuracy.

Main Methods:

• Manual tracing of the intracranial cavity was performed as a reference standard for ICV measurement.
• Four automated software packages (AFNI, Freesurfer, FSL, SPM) were used to estimate ICV.
• Linear regression analyses were conducted to compare automated estimates against reference measurements across five distinct subject groups.

Main Results:

• SPM showed the best performance for adult controls (R(2)=0.67, p<0.01).
• Freesurfer was most accurate for adults with dementia (R(2)=0.46, p=0.02).
• AFNI demonstrated superior accuracy for pediatric controls (R(2)=0.97, p<0.01), while FSL performed best for pediatric epilepsy groups (R(2)=0.6, p<0.01).

Conclusions:

• The optimal automated software for ICV estimation is population-dependent.
• Atlas-based versus non-atlas-based software approaches may influence accuracy across different demographic and clinical groups.
• Software selection for ICV estimation requires careful consideration of the specific study population and neuroimaging data.