This paper presents a method for semi-supervised MAP (maximum a-posterior probability) segmentation of brain tissues where labelled data are available for either all types of tissues or only a few types of tissues possibly at different levels of quality. The proposed MAP segmentation takes supervised and unsupervised segmentation as its two special cases where, respectively, quality labelled data is available or there is no labelled data at all. Experiments on real MR images have shown that the proposed method improved the segmentation accuracy substantially with only a few labelled data in comparison with both fully supervised method with the same labelled data set and unsupervised method.
This paper originally appeared as: Li, W, deSilver, C and Attikiouzel, Y, A semi-supervised map segmentation of brain tissues, Proceedings. ICSP '04. 2004 7th International Conference on Signal Processing, 31 August - 4 September 2004, vol 1, 757-760. Copyright IEEE 2004.