Comparative Analysis of Four Commercially Available Artificial Intelligence Software for Brain Volumetry

Area of Science:

• Neuroimaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Automated brain volumetry is crucial for neurological disorder assessment.
• Variability between different software tools can limit the interchangeability of absolute volume measures.
• Standardization is needed to ensure reliable interpretation of volumetric data.

Purpose of the Study:

• To evaluate inter-software variability and agreement among four AI-based brain volumetry tools.
• To assess the interchangeability of absolute brain volume measurements across different platforms.
• To compare volumetric outputs for various brain regions including lobes, hippocampus, and ventricles.

Main Methods:

• Retrospective comparison of four clinically available AI tools (Neurophet AQUA®, icobrain dm®, mdbrain®, Pixyl®).
• Analysis of identical 3D T1-weighted MRI scans from 47 adults.
• Extraction of absolute volumes for specific brain structures and calculation of intraclass correlation coefficients (ICC) to assess agreement.

Main Results:

• Significant method-dependent scale offsets (20-40%) were observed across all regions.
• Inter-software agreement varied by structure, with excellent agreement for lateral ventricles (avg-ICC ≈ 0.98) and moderate-to-good for hippocampus and total brain volume (avg-ICC ≈ 0.79-0.81).
• Frontal lobe analysis showed the lowest agreement (avg-ICC ≈ 0.75).

Conclusions:

• Absolute volumetric outputs from different AI platforms are not directly interchangeable.
• Relative subject rankings are largely maintained, supporting platform-consistent analysis for longitudinal studies.
• Cross-site studies require data calibration or harmonization before pooling results.