Optimal feature subset selection for MRI brain tumor classification using improved ant-lion optimization

An accurate diagnosis of the type of brain tumor is crucial for experts to choose the best course of therapy and increase the patient’s lifetime. The manual categorization of brain tumors in Magnetic resonance imaging (MRI) with similar structures or appearances can be complex and relies on expertise. Also, brain MRI contains a significant number of redundant or irrelevant features, and classification algorithms struggle to detect patterns without feature selection. By selecting prominent features, the machine-learning model can produce a high level of predicted accuracy in medical diagnostic problems. In this paper, an improved ant lion optimization approach (IALO) is developed to select the optimal features for wrapper-mode classification. Initially, pre-processing is carried out to remove bias and noise. Tumor segmentation is performed using active contours driven by Laplacian of Gaussian energy and optimal region scalable fitting energy. From the segmented brain images, hand-crafted features are extracted using the Histogram of Oriented Gradient, Gray level co-occurrence matrix, Gray level run length matrix, and deep features are extracted using the ResNet50. The proposed IALO is employed for the selection of the best features from handcrafted and deep features. The final step was to classify brain tumors using Support Vector Machines, Naive Bayes, and Recurrent Neural Networks. The proposed IALO-based RNN exhibited enhanced accuracy of 99.59% for BRATS and 99.18% for J.Cheng brain images respectively. From the experimental results, it is observed that our proposed model outperforms the existing metaheuristic optimization algorithms and effectively removes the least significant features while improving classification accuracy.