K-Means clustering and neural network for object detecting and identifying abnormality of brain tumor

• Published in: Soft computing (Berlin, Germany) Germany), 2019-10, Vol.23 (19), p.9083-9096
• Main Authors: Arunkumar, N., Mohammed, Mazin Abed, Abd Ghani, Mohd Khanapi, Ibrahim, Dheyaa Ahmed, Abdulhay, Enas, Ramirez-Gonzalez, Gustavo, de Albuquerque, Victor Hugo C.
• Format: Article
• Language: English
• Subjects: Artificial Intelligence; Brain; Brain cancer; Brain research; Classification; Cluster analysis; Clustering; Computational Intelligence; Control; Engineering; Focus; Fuzzy logic; Fuzzy sets; Gray scale; Magnetic resonance imaging; Mathematical Logic and Foundations; Mechatronics; Medical imaging; Methods; Neural networks; Neuroimaging; Robotics; Tumors; Vector quantization
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-018-3618-7

Brain tumor diagnosis is a challenging and difficult process in view of the assortment of conceivable shapes, regions, and image intensities. The pathological detection and identification of brain tumor and comparison among normal and abnormal tissues need grouped scientific techniques for features extraction, displaying, and measurement of the disease images. Our study shows an improved automated brain tumor segmentation and identification approach using ANN from MR images without human mediation by applying the best attributes toward preparatory brain tumor case revelation. To obtain the exact district region of brain tumor from MR images, we propose a brain tumor segmentation technique that has three noteworthy improvement focuses. To begin with, K -means clustering will be utilized as a part of the principal organization in the process of improving the MR image to be marked in the districts regions in light of their gray scale. Second, ANN is utilized to choose the correct object in view of training phase. Third, texture feature of brain tumor area will be extracted to the division stage. With respect to the brain tumor identification, the grayscale features are utilized to analyze and diagnose the brain tumor to differentiate the benign and malignant cases. According to the study results demonstrated that: (1) enhancement adaptive strategy was utilized as post-processing in brain tumor identification; (2) identify and build an assessment foundation of automated segmentation and identification for brain tumor cases; (3) highlight the methods based on region growing method and K -means clustering technique to select the best region; and (4) evaluate the proficiency of the foreseen outcomes by comparing ANN and SVM segmentation outcomes, and brain tumor cases classification. The ANN approach classifier recorded accuracy of 94.07% with line assumption (brain tumor cases classification) and sensitivity of 90.09% and specificity of 96.78%.