Unsupervised segmentation for multiple sclerosis lesions in multimodality Magnetic Resonance images

Ziming Zeng, Siping Chen, Lidong Yin, Reyer Zwiggelaar

Research output: Chapter in Book/Report/Conference proceedingOther chapter contribution

3 Citations (SciVal)

Abstract

In this paper, a new unsupervised approach is proposed for the segmentation of Multiple Sclerosis (MS) lesions in multimodality Magnetic Resonance (MR) images. The proposed segmentation scheme is based on joint histogram modelling followed by false positive reduction and alpha matting, which is used to deal with the tissue density overlap problem and partial volume effects in MR images. Firstly, the joint histogram is generated by using fluid-attenuated inversion recovery (Flair), T1-weighted (T1-w) and T2-weighted (T2-w) MRI. Then the region for MS lesions in the joint histogram are located. Sub-sequently, the located region is projected back into the 2D MR images with potential MS lesions. Secondly, priori information is utilized to remove false positive volume of interests. Finally, the partial volume effect is modelled by using an alpha technique provides region level lesion refinement. Validation is performed on real multi-channel T1-w, T2-w, and Flair MR volumes. The experimental results show the proposed method can obtain better results than some state-of-the-art methods
Original languageEnglish
Title of host publication2015 8th International Conference on BioMedical Engineering and Informatics (BMEI 2015)
PublisherIEEE Press
Pages126-130
DOIs
Publication statusPublished - 2015

Fingerprint

Dive into the research topics of 'Unsupervised segmentation for multiple sclerosis lesions in multimodality Magnetic Resonance images'. Together they form a unique fingerprint.

Cite this