Article Text

PDF
THU0160 White matter lesions are predominantly demyelinating in systemic lupus erythematosus. An support vector machines classification of texture parameters
  1. A.T. Lapa1,
  2. M.P. Bento1,
  3. L. Rittner1,
  4. H.H. Ruocco1,
  5. G. Castellano1,
  6. B.D. Damasceno1,
  7. L.T.L. Costallat1,
  8. S. Appenzeller2,
  9. R. Lotufo1,
  10. F. Cendes1
  1. 1State University of Campinas
  2. 2Medicine, State UNiversity of Campinas, Campinas, Brazil

Abstract

Background Texture analysis (TA) is a branch of image processing which seeks to reduce image information by extracting texture descriptors from the image. White matter hyperintensities (WMH) are frequently observed in systemic lupus erythematosus (SLE), however the etiology is still unknown. Ischemic and demyelination have been proposed as possible etiologies. Support vector machines (SVM) are a group of supervised learning methods that can be applied to classification or regression.

Objectives To develop a classifier based on neural network to identify the etiology of WMH in SLE.

Methods TA was applied to axial T2-weighted magnetic resonance images (MRI) of 30 SLE, 30 MS, and 30 stroke patients and 30 normal age and sex-matched controls. The TA approach used was based on the Gray Level Co-occurrence Matrices (GLCM).The WMH were manually segmented for each subject, classified in periventricular and subcortical WMH and 256 texture parameters were computed for each lesion. A SVM classifier was developed based on texture features of normal white matter and WMH in MS and stroke patients. The classifier was then used to classify WMHin SLE patients. Nature of the classified WMH, demographic, clinical and laboratory features were included in a regression model to determine which variables could support the possible nature of WMH in clinical practice.

Results We achieve an accuracy rate of 0.93 to classify normal white matter and WMH in MS and stroke patients using SVM technique. Of the 37 periventricular WMH, 29 (78%) were classified as demyelination, 4 (11%) as ischemic and 4 (11%) as normal white matter. Of 53 subcortical lesions, 26 (72%) were classified as demyelination, 6 (11%) as ischemic and 9 (17%) as normal white matter. Age (odds ratio [OR] 1.7, 95% confidence interval ([95% CI] 1.58-6.72), hypertension (OR=2.6; 95%CI 1.9-5.3) and positive antiphospholipid antibodies (aPL) (OR=1.9; 95%CI 1.2-7.3) were variables associated with stroke, whereas shorter disease duration (OR=3.1; 95%CI 2.2-7.5) and new onset of neurologic symptoms (OR=1.8; 95%CI 1.2-3.5) were associated with demyelination.

Conclusions Although 75% of WMH were classified as demyelinating in nature, we identified approximately 25% of ischemic WMH or normal white matter in SLE. SVM of TA is a useful method to help to determine etiology of WMH in SLE. Age, hypertension and aPL were variables associated with ischemic; shorter disease duration and new onset neurologic symptoms were associated with demyelinating lesions in this cohort.

  1. Castellano G, Bonilha L, Li LM, Cendes F. Texture analysis of medical images. Clin Radiol. 2004;59:1061-1069.

  2. Appenzeller S, Vasconcelos FA, Li LM, Costallat LT, Cendes F. Quantitative magnetic resonance imaging analyses and clinical significance of hyperintense white matter lesions in systemic lupus erythematosus patients. Ann Neurol. 2008;64:635-643

Disclosure of Interest A. Lapa Grant/Research support from: FAPESP 2010/13639-1, M. Bento: None Declared, L. Rittner: None Declared, H. Ruocco: None Declared, G. Castellano: None Declared, B. Damasceno: None Declared, L. Costallat: None Declared, S. Appenzeller Grant/Research support from: FAPESP 2008/02917-0; Conselho Nacional Pesquisa Desenvolvimento-Brasil CNPq (300447/2009-4), R. Lotufo: None Declared, F. Cendes: None Declared

Statistics from Altmetric.com

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.