Tumor Localization and Classification from MRI of Brain using Deep Convolution Neural Network and Salp Swarm Algorithm

• Published in: Cognitive computation Vol. 16; no. 4; pp. 2036 - 2046
• Main Authors: Alyami, Jaber, Rehman, Amjad, Almutairi, Fahad, Fayyaz, Abdul Muiz, Roy, Sudipta, Saba, Tanzila, Alkhurim, Alhassan
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.07.2024; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Artificial neural networks; Brain; Brain cancer; Brain research; Classification; Computation by Abstract Devices; Computational Biology/Bioinformatics; Computer Science; Data augmentation; Datasets; Deep learning; Diagnosis; Effectiveness; Feature extraction; Feature selection; Machine learning; Magnetic resonance imaging; Regression analysis; Tomography; Tumors; Brain tumor; Deep learning; MRI; Transfer learning; Public health; Health risks
• ISSN: 1866-9956; 1866-9964
• DOI: 10.1007/s12559-022-10096-2

Early diagnosis of brain tumors is crucial for treatment planning and increasing the survival rates of infected patients. In fact, brain tumors exist in a range of different forms, sizes, and features, as well as treatment choices. One of the essential roles of neurologists and radiologists is the diagnosis of brain tumors in their early stages. However, manual brain tumor diagnosis is difficult, time-consuming, and prone to error. Based on the problem highlighted, an automated brain tumor detection system is mandatory to identify the tumor in its initial stages. This research presents an efficient deep learning-based system for the classification of brain tumors from brain MRI using the deep convolutional network and salp swarm algorithm. All experiments are performed using the publicly available brain tumor Kaggle dataset. To enhance the classification rate, preprocessing and data augmentation such as skewed data ideas are devised. In addition, AlexNet and VGG19 are leveraged to perform specific functionality. Finally, all features merged into a single feature vector for brain tumor classification. Some of the extracted features found insignificant towards effective classification. Hence, we employed an efficient feature selection technique named slap swarm to find the most discriminative features to attain best tumor classification rate. Finally, several SVM kernels are merged for the final classification and 99.1% accuracy is achieved by selecting 4111 optimal features from 8192.