Analyzing Brain Tumor from Structural MR Images Using Kernel Support Vector Machine and Principal Component Analysis

• Published in: SN computer science 2023-11, Vol.4 (6), p.726, Article 726
• Main Authors: Yugander, P., Jagannath, M., Nathiya, N.
• Format: Article
• Language: English
• Subjects: Abnormalities; Algorithms; Approximation; Brain; Brain cancer; Classification; Computer Imaging; Computer Science; Computer Systems Organization and Communication Networks; Data Structures and Information Theory; Datasets; Decision trees; Decomposition; Discrete Wavelet Transform; Imaging techniques; Information Systems and Communication Service; Kernel functions; Machine learning; Magnetic resonance imaging; Mathematical analysis; Medical imaging; Neuroimaging; Original Research; Pattern Recognition and Graphics; Polynomials; Principal components analysis; Sensitivity; Software Engineering/Programming and Operating Systems; Support vector machines; Tumors; Vision; Wavelet transforms
• ISSN: 2661-8907; 2662-995X; 2661-8907
• DOI: 10.1007/s42979-023-02208-y

Accurate and automated classification of magmatic resonance (MR) brain image is a vital task for medical image interpretation and analysis. This paper proposed a new method for brain tumor detection with the two-dimensional discrete wavelet transform (2D-DWT), kernel support vector machine (KSVM), and principal component analysis (PCA). The abnormalities of the human brain cannot be identified using traditional imaging techniques. Human brain neural architectures can be distinguished and classified using MR imaging techniques. In this paper, structural MR Images are used to classify brain tumors. The proposed algorithm is divided into three stages: pre-processing, classification, and post-processing. In pre-processing stage, 2D-DWT and PCA were employed to obtain the MR image features, and PCA is used to decrease the size of the feature vector. The KSVM is used to classify benign tumors from structural MR images. This employs three types of kernel functions: linear, polynomial, and Gaussian radial basis (GRB) kernels. In the last stage stratified K-fold cross-validation is used to avoid the overfitting problem. The proposed algorithm was performed on 80 structural MR images using three different kennel functions. Among the three kernels, the GRB kernel achieved the highest performance than linear and polynomial kernels in terms of accuracy, sensitivity, and specificity. The GRB kernel achieved an accuracy of 98.75%, sensitivity of 98%, and specificity of 99%.