A new brain tumor diagnostic model: Selection of textural feature extraction algorithms and convolution neural network features with optimization algorithms

• Published in: Computers in biology and medicine Vol. 148; p. 105857
• Main Author: Başaran, Erdal
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.09.2022; Elsevier Limited
• Subjects: Accuracy; Algorithms; Artificial neural networks; Biopsy; Brain cancer; Brain damage; Brain injury; Brain tumor; Brain tumors; Convolutional neural network; Datasets; Deep learning; Diagnostic systems; Feature extraction; Feature selection; Genetic algorithms; Heuristic methods; Imaging techniques; Internal Medicine; Machine learning; Magnetic resonance imaging; Medical diagnosis; Medical imaging; Methods; Model accuracy; Neural networks; Neuroimaging; Optimization algorithms; Other; Particle swarm optimization; Support vector machines; Textural feature exraction; Tumors; Wavelet transforms; Brain tumor; Optimization algorithms; Feature selection; Textural feature exraction; Convolutional neural network
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2022.105857

Brain tumors are one of the most dangerous diseases that affect human health and maybe result in death. Detection of brain tumors can be made by using biopsy. However, this is an invasive procedure. It is an extremely dangerous procedure because it can cause bleeding and damage certain brain functions. For this reason, the type and the stage of the disease can be determined after a detailed examination by medical imaging techniques made by field experts. In this study, a computer-based hybrid diagnostic model with high accuracy rate is proposed to diagnose normal brain and brain having types of tumors from brain images obtained by magnetic resonance imaging (MRI) techniques. This diagnostic model consists of three stages. In the first stage, the features of the images were obtained with two different traditional methods, which are widely used in the literature, and the results were examined. In the second stage, different convolutional neural networks that can learn comprehensive information about images were used and the results were tested by obtaining the features of the images. In the third stage, all the feature sets that are obtained were combined, and genetic algorithms, particle swarm optimization technique and artificial bee colony optimization techniques were used for feature selection. The common features of the optimization techniques were used only once. Thus, metaheuristic optimization algorithms were used for feature selection and distinctive features of the images appeared. Feature sets were classified using support vector machine kernels. The proposed diagnostic model is better than the directly used methods with an accuracy rate of 98.22%. Consequently, this method can also be used in clinic service to diagnose tumor by using images of brain MRI. [Display omitted]