Dual feature extraction based convolutional neural network classifier for magnetic resonance imaging tumor detection using U-Net and three-dimensional convolutional neural network. (July 2022)

• Record Type: Journal Article
• Authors: Suresh Kumar, R.; Nagaraj, B.; Manimegalai, P.; Ajay, P.
• Is Part Of: Computers & electrical engineering. Volume 101(2022)
• Publication Date: 2022-07
• Subjects: Medical imaging -- Segmentation -- U-net -- 3DCNN -- Brain tumor; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2022.108010
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: Gray-Level Co-occurrence matrix and vantage point tree were used which has excellent performance and classifier. Networks with two segmentation such as U-Net and 3D CNN are proposed for better and accurate predictions. 3D CNN was used to segment the enhancing tumor more accurately. U-Net was used in the tumor core for the better performance used in the sub-region of final prediction. Abstract: Analysis and monitoring of disease development rely heavily on automated segmentation of brain tumors using MRI data. Because gliomas are aggressive and diverse, effective and precise segmentation techniques are utilized to divide tumors into intra-tumoral groups. In the proposed work, the Gray Level Co-occurrence Matrix feature extraction is used to extract characteristics from the image. In the segmentation problem of Brain tumor, Convolutional Neural Networks, which are widely employed in biomedical image segmentation, have greatly enhanced the state-of-the-art accuracy. In this work, we propose a major but simple combinative approach that results in improved and more precise estimates by combining two segmentation networks: a U-Net and a 3D CNN. For each model, it was assessed independently on the dataset to produce segmentation maps that varied greatly in terms of segmented tumor sub-regions and were then ensembled to reach the final prediction. On the validation set, the accuracy (percentage) of 98.29, 98.45, and 99.4 for tumor core, enhanced tumor and whole tumor, respectively, performed well in contrast to state-of-the-art designs presently existing. Graphical abstract: Image, graphical abstract