A Multi-Texture Based Feature Vector for Classification of MR Images of Patients Suffering from Dementia

Dementia is a prevalent neurological disorder affecting millions globally which poses significant challenges in early detection and classification. In this research work, a hybrid texture based feature extraction method is presented for classification of Magnetic Resonance Imaging brain images having dementia using machine learning techniques. The study investigates diverse methodologies, including extensive image pre-processing and feature extraction methods based on multiple textures to enhance structural details crucial for effective analysis. Results indicate that while Support Vector Classifier exhibits a moderate performance with 50.4% accuracy, Gradient Boosting Classifier and Random Forest Classifier showcase commendable accuracies of 92.7% and 99.0%, respectively. Remarkably, Simple Neural Network Classifier outperforms all classifiers with a remarkable accuracy of 99.7%. Comparative analysis highlight Simple Neural Network's superior capability, emphasizing its potential as a reliable tool for precise image classification is also presented.