An improved hybrid classification of brain tumor MRI images based on conglomeration feature extraction techniques

• Published in: Neural computing & applications Vol. 34; no. 11; pp. 9069 - 9086
• Main Authors: Bansal, Twinkle, Jindal, Neeru
• Format: Journal Article
• Language: English
• Published: London Springer London 01.06.2022; Springer Nature B.V
• Subjects: Accuracy; Artificial Intelligence; Brain; Classifiers; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Datasets; Decision trees; Feature extraction; Image classification; Image Processing and Computer Vision; Image segmentation; K-nearest neighbors algorithm; Magnetic resonance imaging; Nervous system; Original Article; Parameter sensitivity; Probability and Statistics in Computer Science; Tumors; Gray level co-occurrence matrix; Conglomerate classifier; Random forest; Hybridization; Marker-controlled watershed; Decision tree
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-022-06929-8

The brain seems to be the most complex organ in the human body and operates as the central part of the nervous system. The classification of brain tumors is the most complicated problem which medical experts face. In our proposed work, the novelty is that we have adopted the hybridization of feature exaction techniques and a conglomerate of classification methods. The hybrid features are extracted from the tumor region using the Gray Level Co-occurrence Matrix technique and shape-based features. There are three different types of brain magnetic resonance imaging datasets used in work. For the segmentation, the Marker-controlled watershed segmentation strategy is implemented in this work. In this work, the hybrid classifier, which includes the Random Forest (RF) classifier, K-Nearest Neighbour classifier, and Decision Tree classifier, is implemented. Dataset 1 (China Hospital data) shows a better overall accuracy of 98.75%; for dataset 2, accuracy is 98.99%; and for dataset 3, the achieved accuracy is 98.92%. The other parameters that prove the overall work's efficacy include sensitivity, precision, specificity, F1-score, Cohen’s kappa value, Dice coefficient, Jaccard Index, and area under the curve (AUC). The proposed work is also compared with the state-of-the-art methods and observed better one.