Brain Tumor Detection and Classification Using a Hybrid Convolutional Neural Network Learning Method with Support Vector Machine (CNN-SVM) Based on Fuzzy Weighting and Transit Search Optimization (TSO) Algorithm

The diagnosis and categorization of brain tumors, which are typically based on invasive histological evaluation of biopsy samples, are essential for successful treatment. Current techniques are time-consuming, labor-intensive, and prone to errors. Meanwhile, machine learning techniques, including convolutional neural networks (CNNs), frequently struggle with accuracy and high computational complexity, especially when dealing with big 3D MRI datasets. To classify brain tumors into low-grade gliomas (LGG) and high-grade gliomas (HGG) with greater accuracy and less training complexity, this project intends to create a fully automated, non-invasive method. We suggest a hybrid strategy that integrates Takagi–Sugeno (T–S) fuzzy logic systems with 3D Fully Convolutional Neural Networks (3D-FCNN) equipped with Support Vector Machines (3D-FCNN-SVM) or Transit Search Optimization (3D-FCNN-TSO). Convolutional and classification layers are optimized using the TSO method in a two-stage training process, while CNN training parameters are decreased using a fuzzy weighting technique. The Kaggle BraTS dataset was used to train and evaluate the model. In contrast to conventional CNNs, the suggested 3D-FCNN-TSO model significantly reduced training parameters (from 2365 to 269) and increased training speed while achieving a classification accuracy of over 95%. With excellent accuracy and low computing cost, the 3D-FCNN-TSO and 3D-FCNN-SVM models provide a reliable and effective method for classifying brain tumors, potentially leading to practical therapeutic applications.