Available online at www.sciencedirect.com Computerized Medical Imaging and Graphics 32 (2008) 124–133 A novel method for automatic determination of different stages of multiple sclerosis lesions in brain MR FLAIR images Rasoul Khayati a , Mansur Vafadust a , Farzad Towhidkhah a,∗ , S. Massood Nabavi b a Biomedical Engineering Faculty, Amirkabir University of Technology, Tehran, Iran b Neurology Department, Medical Faculty, Shahed University, Tehran, Iran Received 8 February 2007; received in revised form 17 October 2007; accepted 18 October 2007 Abstract It is very important to detect stages of multiple sclerosis (MS) lesions in order to exactly quantify involved voxels. In this paper, a novel method is proposed for automatic detection of different stages of MS lesions in the brain magnetic resonance (MR) images, in fluid attenuated inversion recovery (FLAIR) studies. In the proposed method, firstly, MS lesion voxels are segmented in FLAIR images based on adaptive mixtures method (AMM) and Markov Random Field (MRF) model. Then, signal intensity of each lesion voxel is modeled as a linear combination of signals related to the normal and also abnormal parts, in the voxel. By applying an optimal threshold, voxels with new intensities are primarily classified into two stages: previously destructed (chronic) and on going destruction (acute) lesions. Finally, the acute lesions, according to their activities, are classified, by another optimal threshold, into two new stages, early and recent acute. Evaluation of the proposed method was performed by manual segmentation of chronic and enhanced (early) acute lesions in gadolinium enhanced T1-weighted (Gad-E-T1-w) images by studying T1-weighted (T1-w) and T2-weighted (T2-w) images, using similarity criteria. The results showed a good correlation between the lesions segmented by the proposed method and by experts manually. Thus, the suggested method is useful to reduce the need for paramagnetic materials in contrast enhanced MR imaging which is a routine procedure for separation of acute and chronic lesions. © 2007 Elsevier Ltd. All rights reserved. Keywords: Multiple sclerosis lesions; Staging; MRI 1. Introduction MS is one of the progressive central nervous system diseases, which causes morphological and structural changes to the brain. Quantitative assessment of the changes in the brain MR images in association with clinical judgment can provide more accu- rate assessment of the disease progress. Also, it can provide important information in order to find out the most effective therapeutic method for patients. Due to high resolution and good differentiation between soft tissues and other structures, MRI has superiority to the other imaging techniques for the studies of nervous system diseases. MRI has been known as the best par- aclinical examination for MS which can reveal abnormalities in 95% of the patients [1]. Because of inflammation and destruc- ∗ Corresponding author. Tel.: +98 21 64542363; fax: +98 21 66495655. E-mail addresses: Towhidkhah@aut.ac.ir, Towhidkhah@gmail.com (F. Towhidkhah). tion of myelin in central nervous and subsequent pathological changes, MR images of MS patients show scattered singular or collective plaques with small or big sizes in different shapes. The appearances of the lesions in different MRI images including, T1-w, T2-w and FLAIR are not the same (see Fig. 1). There are many proposed approaches for segmentation of the brain tissues as well as automatic and semi-automatic detection methods of MS lesions according to different degrees of automa- tion or user interference [2–10]. Recently, Anbeek et al. [11] have proposed a novel automatic approach for segmentation of white matter lesions in the MR images of brain. The introduced algorithm uses different information, including voxel intensity and the spatial information, to classify voxels by a K-Nearest Neighbor (KNN) classifier. This technique assigns a probability to each voxel for being part of white matter lesion. Similarity Index (SI), then, is used for determination of the optimal thresh- old on the probability map, to segment the images. They have showed the high accuracy of their approach, in comparison with the other methods for similar task. These methods have been 0895-6111/$ – see front matter © 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.compmedimag.2007.10.003