Multiple Sclerosis Slice Identification by Haar Wavelet Transform and Logistic Regression

Abstract

Abstract. (Aim) Currently, scholars tend to use computer vision approaches to implement multiple sclerosis (MS) identification. (Method) In this study, we proposed a novel MS slice identification system, based on Haar wavelet transform, principal component analysis, and logistic regression. (Result) Simulation results showed the accuracies of our method using 2-level, 3-level, and 4-level decomposition are 83.25±1.62%, 89.72±1.18%, and 87.65±1.79%, respectively. (Conclusion) Our method with 3-level decomposition achieved the best. BackgroundMultiple sclerosis (MS) is a progressive disease that affects both brain and spinal cord [1]. Early symptoms are composed of tingling, weakness, blurred version, numbness, etc. [2]. The traditional MRI scanning based identification may meet with the "normal-appearing white matter" problem. Hence, it is necessary to develop new methods to identify MS plaques.Traditionally, scholars have proposed many computer vision methods [3][4][5] in identify abnormal brain diseases [6][7][8][9][10][11][12][13][14][15][16][17][18][19][20], including MS slice identification. For example, Zhou (2016) [21].used stationary wavelet entropy, combined with several machine learning methods. Zhan and Chen (2016) [22] used biorthogonal wavelet transform and logistic regression. Karaca (2017) [23] used convex combination of infinite kernels.Nevertheless, those methods are too complicated, and their models are difficult to train. In this study, we presented a novel and simple system, which was based on Haar wavelet transform, principle component, and logistic regression. Materials and MethodsThirty-four multiple sclerosis patients and thirty-three healthy subjects were enrolled from China local hospitals from 2013. Experienced radiologists were instructed to select the slice with MS plaques from the 34 patients, and select corresponding slices from 33 healthy subjects. In total, we selected 141 slices from MS patients and 148 slices from healthy controls.Our method contains three-stages. In the first stage, Haar wavelet [24] was chosen to transform the brain slices from spatial domain to wavelet domain [25][26][27][28][29][30][31]. The mother wavelet Haar wavelet λ(b) is defined with the form ofHere b denotes the spatial axis: either horizontal or vertical. The scaling function θ(b) is defined as:International Conference on Advances in Materials, Machinery, Electrical Engineering (AMMEE 2017) Copyright © 2017, the Authors. Published by Atlantis Press. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). Advances in Engineering Research, volume 114We chose to use a two-level Haar wavelet transform. In the second stage, principle component was used to reduce the wavelet coefficients to only cover the threshold (THR) of total variances. We here set THR to 95%.In the third stage, logistic regression (LR) [32] was employed to be the classifier. The logistic regression is simpler and easier to train than multilayer perceptron [33][34][35][36], probabilis...

Multiple Sclerosis Identification by 14-Layer Convolutional Neural Network With Batch Normalization, Dropout, and Stochastic Pooling

“…In this experiment, we compared our CNN-DO-BN-SP method with traditional AI methods: Multiscale AM-FM (Murray et al, 2010 ), ARF (Nayak et al, 2016 ), BWT-LR (Wang et al, 2016 ), 4-level HWT (Wu and Lopez, 2017 ), and MBD (Zhang et al, 2017 ). The results were presented in Table 10 .…”

“…Wang et al ( 2016 ) combined biorthogonal wavelet transform (BWT) and logistic regression (LR). Wu and Lopez ( 2017 ) used four-level Haar wavelet transform (HWT). Zhang et al ( 2017 ) proposed a novel MS identification system based on Minkowski-Bouligand Dimension (MBD).…”

Automatic detection of multiple sclerosis lesions using Mask R‐CNN on magnetic resonance scans

“…Zhang et al [34] compared the detection of MS lesions on a dataset using nearest neighbour algorithm, support vector machines and decision tree. In another study, Wu and Lopez performed lesion detection in MR slices using Haar wavelet transforms, principal component analysis and logistic regression [35].…”

Investigation of Deep-Learning-Driven Identification of Multiple Sclerosis Patients Based on Susceptibility-Weighted Images Using Relevance Analysis

“…Recently, a few studies have focused on MS classification based on convolutional neural networks (CNNs) without lesion segmentation (Wang et al, 2018;Zhang et al, 2018;Marzullo et al, 2019). Zhang et al (2018) have proposed a 10-layer CNN-PreLU-Dropout approach for identifying MS patients based on 2D T 2 -weighted axial MRI data that outperforms other modern MS identification approaches (Murray et al, 2010;Wang et al, 2016;Wu and Lopez, 2017;Ghirbi et al, 2018). Wang et al (2018) have proposed an improved structure of the CNN-PreLU-Dropout approach (Zhang et al, 2018) by incorporating batch normalization, and stochastic pooling applied to the same data and achieved superior performance compared to the original method (Zhang et al, 2018).…”