Efficient Feature Extraction for Classification of Alzheimer's Abnormalities Using Deep Learning

Recent machine learning AD predictions employed feature selection and extraction methods. The literature research revealed several attempts to identify AD by using different Deep learning layer and micro-simulation techniques. However, identifying important features that can detect AD extremely early remains a difficult task. Automation, poor classification accuracy, over fitting, etc. are the other challenges while using machine learning algorithms. The recent progression of using deep learning methods has received the significant attention for the automatic AD classification. CNN's strategy, while successful, has certain drawbacks: (1) The CNN model uses local receptive fields, shares weights, and subsamples, (2) deep neural networks need a lot of labeled sample training models to avoid over fitting, which may cause the loss of crucial information like representative features at various frequencies and orientations; and (3) the higher computational overhead is another challenge. Therefore, these challenges motivate us the need of the proposed study. The scope of the proposed study is limited to the classification of AD from the input medical images.