A comparison of MR based segmentation methods for measuring brain atrophy progression.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Accurate automated brain segmentation is vital for clinical applications detecting subtle brain volume changes over time.
• Established methods like SIENA, US, and kNN require comparative evaluation for precision and accuracy using ground-truth data.

Purpose of the Study:

• To compare the precision (repeatability) and accuracy (ground-truth) of SIENA, US, and kNN brain segmentation methods for measuring brain volume change.
• To evaluate the performance of these methods on 1.5 T MRI scans from 10 subjects with one baseline and two follow-up scans over 4 years.

Main Methods:

• Comparison of SIENA, US, and kNN automated brain segmentation techniques.
• Evaluation of precision using coefficient of repeatability for brain volume and volume change.
• Assessment of accuracy through correlation with manual segmentations (Spearman's correlation).

Main Results:

• US exhibited the largest coefficient of repeatability for volume change (2.84%), while kNN (0.31%) and SIENA (-0.92%) showed better precision.
• US and kNN demonstrated good correlation with manual segmentations for absolute brain volume (ρ≥0.96).
• SIENA showed the best correlation for brain volume changes (ρ=0.82), followed by kNN (ρ=0.60) and US (ρ=0.50).

Conclusions:

• US and kNN offer good precision, accuracy, and comparability for absolute brain volume measurements.
• SIENA demonstrates superior performance for measuring brain volume changes.
• kNN serves as a viable alternative for assessing volume changes in other brain structures.