A Novel Classification Method using Effective Neural Network and Quantitative Magnetization Transfer Imaging of Brain White Matter in Relapsing Remitting Multiple Sclerosis

Abstract

Background: Quantitative Magnetization Transfer Imaging (QMTI) is often used to quantify the myelin content in multiple sclerosis (MS) lesions and normal appearing brain tissues. Also, automated classifiers such as artificial neural networks (ANNs) can significantly improve the identification and classification processes of MS clinical datasets.Objective: We classified patients with relapsing-remitting multiple sclerosis (RRMS) from healthy subjects using QMTI and T1 longitudinal relaxation time data of brain white matter and the performance of three ANN-based classifiers have been investigated. Materials and Methods: Conventional magnetic resonance imaging (MRI) and quantitative magnetization transfer scans were obtained from RRMS patients (n=30) and age-matched healthy subjects (n=30). After image pre-processing and brain tissue segmentation, QMTI parameters including magnetization transfer ratio (MTR), magnetization transfer rate (Ksat), T1 relaxation time under MT saturation pulse (T1sat) and T1 longitudinal relaxation time were calculated as parametric maps. Three ANN algorithms, including multilayer perceptron (MLP), radial basis function (RBF) and ensemble neural network based on Akaike information criterion (ENN-AIC) input features were extracted in the form of QMTI and T1 mean values. The ANNs quantitative performance is measured by the standard evaluation of confusion matrix criteria.Results: The results indicate that ENN-AIC-based classification method has achieved 90% accuracy, 92% sensitivity and 86% precision compared to other ANN classification models such as RBF and MLP. NPV, FPR and FDR values of the proposed ENN-AIC model were found to be 0.933, 0.125 and 0.133, respectively. A graphical representation of how to track actual data by the predictive values derived from each of the three algorithms, was also presented. It has been demonstrated that ENN-AIC as an effective neural network improves the quality of classification results compared to MLP and RBF.Conclusion: The efficiency and robustness of ENN classifier will greatly enhance with the use of AIC-based combination weights assignment. In addition, this research provides a new direction to classify a large amount of quantitative MRI data that can help the physician in a correct MS diagnosis.

Review of advanced computational approaches on multiple sclerosis segmentation and classification

“…7. This article reviewed more than 70 state- [54] non-local means (NLM) NA Brainweb database concept of 5D NLM is used to achieve higher accuracy with lower distortion Valverde et al 2014 [55] FAST and SPM8 NA OASIS database high accuracy WM lesion filling is achieved using FAST and SPM8 techniques Khotanlou and Afrasiabi 2011 [56] spatially constrained possibilistic fuzzy C means (SCPFCM) [37] 2016 differential evolution 81.68 Yeliz, K. and Şengül, H. [63] 2015 feed-forward back propagation 96.75 Esposito et al [40] 2010 evolutionary-fuzzy 88.71 Barry R. Greene et al [41] 2015 logistic regression model 96.90 Ayelet Akselrod-Ballin et al [44] 2009 decision forest 98.50 Simaa Hamid et al [45] 2016 grey level run length matrix 96.90 Yudong Zhang et al [47] 2016 SWE + KNN 97.94 Youngjin Yoo et al [52] 2016 CNN + Euclidean distance transform 72.90 Chang et al [64] 2018 deep CNN 94 Yijun Zhao et al [65] 2017 SVM 81 Punal M. Arabi et al [23] 2017 machine vision 90 Arman Eshaghi et al [66] 2016 random forest 80 Yu-Dong Zhang et al [53] 2018 CNN-PReLU-Dropout 98.23 Zhou and Shen [48] 2018 biogeography-based optimisation with GLCM 92.75 Washimkar and Chede [24] 2017 KNN 97 Rodrigo Antonio Pessini et al [67] 2018 KNN 95 Siar and Teshnehlab [68] 2019 CNN 96.88 Fooladi et al [38] 2018 ENN-AIC 90 Shui-Hua Wang et al [ of-the-art articles to provide broad knowledge of the concepts of MS, and analysed the techniques used for the segmentation and classification of MS diagnoses using MRI and images collected from different databases. Here, the challenges of the lesion segmentation and the classification problem were described and solutions identified.…”

“…The values of FPR, NPV, and FDR were observed to be 0.125, 0.933, and 0.133, individually, as indicated by the proposed ENN-AIC model. It was shown that ENN-AIC as an ENN improved the condition of categorisation outcomes compared with RBF and MLP [38].…”

Quantitative magnetisation transfer imaging in relapsing-remitting multiple sclerosis: a systematic review and meta-analysis

“…The majority of studies were European (k=41/86, Supplementary (k=7) 37,57,58,66,84,90,102 and international (k=8) studies (or >3 test centres). 33,34,55,96,108,110,111,114 The top three study locations were London (k=8), 49,[51][52][53][54]64,73,106 Milan (k=8), 46,48,65,78,92,93,97,99 and Lausanne (k=6).…”

“…The radiofrequency pulse shape was generally Gaussian (k=28), 40,45,61,94,95,98 Sinc-Gaussian (k=5), 35,59,74,82 Fermi (k=4), 57,58,84,90 1-2-1 binomial (k=2), 80,91 and hyperbolic secant pulses (k=1) 71 were also used. Forty studies did not specify pulse shape.…”

“…30,33,35,47,50,51,56-58,67,69,72,80,84,91,112 30,33,35,47,50,51,56-58,67,69,72,74,80,84,91,112 . qMT parameters included the R1free (k=7) 30,33,35,47,50,56,69 or T1free (k=5) 51,58,67,80,91 including under saturation (T1sat, k=2), 57,58 T2free (k=4) 33,35,47,69 and T2bound (k=5), 33,35,47,51,69 kf (k=8) 30,33,56,67,69,72,80,91 including under saturation (ksat, k=2), 57,58 the equilibrium magnetisation of the "bound" pool and the non-ideal inversion of the "free" pool signal (M0f and Sf, respectively, k=2), 30,56 f (k=3), 35,47,51 and F (k=2). 30,33,47,50,56,69,72,112…”