Tissue Segmentation

Seg 1 Seg 2

White (left) and grey (right) matter probabilitiy maps from modified Rusenik algorithm

We will demonstrate two approaches for automatic segmentation of brain tissues, the first based on an extension of the technique proposed in [2], which involves the direct solution of tissue volume estimates from two co-registered NMR images. The extensions make possible fully automated application to a range of clinical scanner sequences. The other is based on a Bayes statistical approach which can work with any single NMR image [3]. In both cases a theoretical framework has been developed to predict the expected statistical repeatability for different NMR scans [1]. Accuracy of per voxel tissue volume measurement, estimated from repeatability, differed in accordance with our theoretical predictions by up to a factor of 4 between scans, with best case performance around 10% of a voxel volume. The theoretical models of repeatability make possible the automatic selection of sequences which optimise segmentation performance. Assumptions of normal tissue may produce systematic differences in estimates of grey and white matter for abnormal variations (ie: invalid analytic assumptions) in individual subjects. However, we believe that even in these cases such techniques can be of real benefit in situations which require assessment of relative change and rapid automated region extraction.

Library toolkit
binary 1
Sun Sparc - Solaris 2
binary 2
Intel x86 - Linux 2
[1] N.A.Thacker, A.Jackson, X.P.Zhu and K.L.Li, `Accuracy of Tissue Volume Estimation in MR Images', proc. MIUA, Leeds ,
137-141, 1998.
[2] E.Vokurka, A. Herwadkar, N.A.Thacker, R.T.Ramsden, A.Jackson, `Using Bayesian Tissue Classification to Improve Accuracy of Acoustic Neuroma Volume Measurement', Proc. CARRS 2000, San Francisco, pp 331-336, June, 2000.
Relevant other work
[2] H.Rusinek, et. al. `Alzheimer's Disease: measuring loss of cerebral grey matter with MR imaging'. Radiology, 178(1), p109-14, 1991.
[3] D.H.Laidlaw, K.W.Fleisher and A.H.Barr, `Partial Volume Bayesian Classification of Material Mixtures in MR Volume Data Using Voxel Histograms.' IEE Trans, Med. Imag. 17, 1, 74-86, 1998.
the end
[Back to software]