Optimal DeepMRSeg based tumor segmentation with GAN for brain tumor classification. (April 2022)

• Record Type: Journal Article
• Authors: Neelima, G.; Chigurukota, Dhanunjaya Rao; Maram, Balajee; Girirajan, B.
• Is Part Of: Biomedical signal processing and control. Volume 74(2022)
• Publication Date: 2022-04
• Subjects: Brain tumor classification -- DeepMRSeg -- Generative Adversial network -- Normalization -- Convolutional neural network; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103537
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The highlights of the article are given below for your kind perusal. Kindly, consider and forward my article for further processes. Proposed SPO-based Optimal DeepMRSeg for segmentation : The proposed SPO-based Optimal DeepMRSeg is adapted for generating the segments considering pre-processed MRI. Here, the training of DeepMRSeg is done using proposed SPO, which is devised by combining SFO and PO for tuning optimal weights. Proposed CAViaR-SPO-based GAN for brain tumor classification : The proposed CAViaR-SPO is devised for determining the tumor from the MRI. Here, the training of GAN is done using proposed CAViaR-SPO, which is devised by combining CAViaR and SPO. Here, the SPO is obtained by combining SFO and PO. The proposed CAViaR-SPO-based GAN provided best performance with highest accuracy of 91.7%, highest segmentation accuracy of 90.0%, highest sensitivity of 92.8%, and highest specificity of 92.5%. Abstract: The accurate and timely treatment of brain tumor is considered as an imperative part in effectual planning of treatment. However, the manual categorization of tumor in Magnetic resonance imaging (MRI) with same structures or appearance is complex that relies on expertise to discover brain tumor. This paper devises an automatic mechanism that can perform the cataloguing of tumor with MRI. The pre-processing is termed as initial measure to normalize intensity. Here, pre-processing is carried out with min-max normalization. The segmentation is performed with Optimal DeepMRSeg strategy, wherein the DeepMRSeg is trained using newly devised sailfish Political Optimizer (SPO) algorithm. The proposed SPO is devised by combining sailfish optimization algorithm (SOA) and Political Optimizer (PO). Then the Convolutional neural network (CNN) features are extracted and data augmentation is performed. The data augmentation, like random translation, randomized left or right flipping, brightness, rotation or adjustment of contrast is done with CNN. Then, the classification is done with Generative Adversial network (GAN), and trained using Conditional Autoregressive Value at Risk-based sailfish political Optimizer (CAViaR-SPO) by combining CAViaR, SOA and PO. The proposed CAViaR-SPO-based GAN offered enhanced performance with elevated accuracy of 91.7%, segmentation accuracy of 90%, sensitivity of 92.8%, and specificity of 92.5%.