Glioma detection based on multi-fractal features of segmented brain MRI by particle swarm optimization techniques

详细信息    查看全文

• 作者：Salim Lahmiri^a ; ^b ; ^{salim.lahmiri.1@ens.etsmtl.ca" class="auth_mail" title="E-mail the corresponding author}
• 关键词：Brain magnetic resonance image ; Glioma ; Particle swarm optimization ; Segmentation ; Spectral distribution ; Generalized hurst exponents ; Support vector machines ; Classification
• 刊名：Biomedical Signal Processing and Control
• 出版年：2017
• 出版时间：January 2017
• 年：2017
• 卷：31
• 期：Complete
• 页码：148-155
• 全文大小：1124 K

文摘

We seek to compare three automated diagnosis systems to detect glioma in brain magnetic resonance images (MRIs). Each glioma diagnosis system is composed of four steps. First, a given particle swarm optimization (PSO) technique is used to segment brain magnetic resonance image (MRI). For instance, brain MRI segmentation is performed either by classical PSO, Darwinian particle swarm optimization (DPSO), or fractional-order DPSO (FODPSO). Second, the directional spectral distribution (DSD) signature of the segmented image is computed. Third, multi-fractals of the computed DSD are estimated by generalized Hurst exponents. Fourth, classification of the obtained multi-fractal features is performed by support vector machine (SVM). The leave-one-out cross-validation method (LOOM) is adopted to assess the classification accuracy of all three glioma detection systems. The experimental results indicate that (1) PSO-DSD-MSA system achieved 98.01% ± 0.07 classification accuracy, 100% sensitivity, and 94.78% ± 0.02 specificity, (2) DPSO-DSD-MSA achieved 98.38% ± 0.01 classification accuracy, 99.5% ± 0.02 sensitivity, and 96.70% ± 0.02 specificity, and (3) the FODPSO-DSD-MSA system reached 99.18% ± 0.01 classification accuracy, 100% sensitivity, and 97.95% ± 0.036 specificity. The overall processing time is less than 5 s. All presented systems achieved higher performances in comparison with previous works. As FODPSO-DSD-MSA accuracy is high and processing time is low, it may be a promising automated glioma diagnosis system in clinical milieu.