Significant LOOP with clustering approach and optimization enabled deep learning classifier for the brain tumor segmentation and classification

• Published in: Multimedia tools and applications 2022, Vol.81 (2), p.2365-2391
• Main Authors: Jemimma, T. A., Raj, Y. Jacob Vetha
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Brain; Brain cancer; Classification; Clustering; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Deep learning; Edema; Feature extraction; Image classification; Image segmentation; Information theory; Machine learning; Magnetic resonance imaging; Medical imaging; Multimedia Information Systems; Optimization; Special Purpose and Application-Based Systems; Tumors; Wavelet transforms
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-021-11591-8

Magnetic resonance images (MRI) is the imperative imaging modality utilized in medical diagnosis tool for detecting brain tumors. The MRI possess the capability to offer detailed information based on anatomical structures of brain. However, the major obstacle in the MRI classification is semantic gap among low-level visual information obtained by the high-level information alleged from clinician and MRI machine. This paper proposes the novel technique, named Chaotic whale cat swarm optimization-enabled Deep Convolutional Neural Network (CWCSO-enabled Deep CNN) for brain tumor classification. Here, pre-processing is employed for removing noise and artifacts contained in image. Moreover, Fractional Probabilistic Fuzzy Clustering is employed for segmentation for identifying the tumor regions. Consequently, the feature extraction is carried out from segmented regions of image using wavelet transform, Empirical Mode Decomposition (EMD), scattering transform, Local Directional Pattern (LDP) and information theoretic measures. In addition, Significant LOOP is newly developed through modifying Significant Local Binary Pattern (SLBP) by LOOP. The extracted features are induced by Deep CNN to determine non-tumor, edema, tumor, and enhanced tumor, which is trained by the proposed CWCSO. Thus, the resulted output of proposed CWCSO-based Deep CNN is employed for brain tumor classification. The proposed model showed improved results with maximal specificity of 98.59%, maximal accuracy of 95.52%, and maximal sensitivity of 97.37%, respectively.