Conventional Features and Machine Learning Approach to Predict Stages of Alzheimer’s Disease Using MRI Images

Millions of rural and urban populations across the continents remain undiagnosed from the slow-poisoning Alzheimer’s disease (AD). There has been no effective cure known to date. Unlike other life-risking diseases, AD remains a crucial concern due to the lack of effective methods to prevent its progression and early detection. Patient ignorance, carelessness, and minimal importance in the progressive stage lead AD to capture and control the neuronal system during the autumn of life. This paper introduces an efficient conventional feature and machine learning (CF-ML) approach for detecting AD in the early stages. The AD dataset acquired from the Kaggle data store consists of magnetic resonance imaging images, including non-dementia (ND) and dementia classes (mild, very mild, and moderate dementia). The CF-ML framework consists of preprocessing (region of interest, resizing, and contrast correction), feature extraction (textural features, global statistical features, and filter-based features), cleaning, normalization, dimension reduction, data partitioning, and classification. The performance of the CF-ML framework is evaluated for dementia and ND classes and a multiclass configuration that includes all four classes. Using the simpler and low-complexity CF-ML approach, superior results were obtained for the binary class, whereas results for multiclass were encouraging. The classification accuracy was 99.76% for the binary configuration and 98.82% for the multiclass configuration.