Objective: Brain tissue segmentation by conventional threshold-based techniques may have limited accuracy and repeatability in older subjects. We present a new multispectral magnetic resonance (MR) image analysis approach for segmenting normal and abnormal brain tissue, including white matter lesions (WMLs).
Methods: We modulated two 1.5T MR sequences in the red/green colour space and calculated the tissue volumes using minimum variance quantisation. We tested it on 14 subjects, mean age 73.3 +/- 10 years, representing the full range of WMLs and atrophy. We compared the results of WML segmentation with those using FLAIR-derived thresholds, examined the effect of sampling location, WML amount and field inhomogeneities, and tested observer reliability and accuracy.
Results: FLAIR-derived thresholds were significantly affected by the location used to derive the threshold (P = 0.0004) and by WML volume (P = 0.0003), and had higher intra-rater variability than the multispectral technique (mean difference +/- SD: 759 +/- 733 versus 69 +/- 326 voxels respectively). The multispectral technique misclassified 16 times fewer WMLs.
Conclusion: Initial testing suggests that the multispectral technique is highly reproducible and accurate with the potential to be applied to routinely collected clinical MRI data.