Feature Extraction Using a Residual Deep Convolutional Neural Network (ResNet-152) and Optimized Feature Dimension Reduction for MRI Brain Tumor Classification

• Published in: Diagnostics (Basel) Vol. 13; no. 4; p. 668
• Main Authors: Athisayamani, Suganya, Antonyswamy, Robert Singh, Sarveshwaran, Velliangiri, Almeshari, Meshari, Alzamil, Yasser, Ravi, Vinayakumar
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 01.02.2023; MDPI
• Subjects: Accuracy; Algorithms; Automation; Brain cancer; Brain research; Brain tumors; Canny algorithm; Cell division; Classification; deep convolutional neural network; Deep learning; Diagnosis; Health aspects; Machine learning; Magnetic resonance imaging; Mathematical optimization; Medical imaging; modified chimp optimization algorithm; Neural networks; softmax classifier; spatial gray level dependence matrix; Technology application; Tumors; X-rays; India; spatial gray level dependence matrix; modified chimp optimization algorithm; deep convolutional neural network; softmax classifier; Canny algorithm
• DOI: 10.3390/diagnostics13040668
• ISSN: 2075-4418

One of the top causes of mortality in people globally is a brain tumor. Today, biopsy is regarded as the cornerstone of cancer diagnosis. However, it faces difficulties, including low sensitivity, hazards during biopsy treatment, and a protracted waiting period for findings. In this context, developing non-invasive and computational methods for identifying and treating brain cancers is crucial. The classification of tumors obtained from an MRI is crucial for making a variety of medical diagnoses. However, MRI analysis typically requires much time. The primary challenge is that the tissues of the brain are comparable. Numerous scientists have created new techniques for identifying and categorizing cancers. However, due to their limitations, the majority of them eventually fail. In that context, this work presents a novel way of classifying multiple types of brain tumors. This work also introduces a segmentation algorithm known as Canny Mayfly. Enhanced chimpanzee optimization algorithm (EChOA) is used to select the features by minimizing the dimension of the retrieved features. ResNet-152 and the softmax classifier are then used to perform the feature classification process. Python is used to carry out the proposed method on the Figshare dataset. The accuracy, specificity, and sensitivity of the proposed cancer classification system are just a few of the characteristics that are used to evaluate its overall performance. According to the final evaluation results, our proposed strategy outperformed, with an accuracy of 98.85%.