Alzheimer's Disease Prediction Using Machine Learning Methodologies

Alzheimer is one of the Neuro-degenerative diseases and the first most usual cause of mental illness in elderly people. The part of brain which gets affected in this disease is hippocampus degeneration. This disease should be detected in early stage only then it is possible to save the lives of the patients. It becomes dangerous and sometimes fatal in case of people of age 65 or above. In accordance with the report from the Alzheimer's Association, nearly, 58 lakh American people are currently living with this disease and by 2050 this range is supposed to rise to closely 14 million. Nowadays, Alzheimer's disease treatments may reduce the rate of decline provisionally. These remedies can occasionally help people with this progressive disease. In the modern era, E-Health systems subjected to expeditious development in communication and computing. Therefore, in this proposed work the classification of Alzheimer's disease is done using machine learning in three approaches. In the first approach, feature extraction was carried out manually. A total of six features were extracted. Out of which, Local Binary Pattern (LBP) feature performed well when compared with others. The highest accuracy obtained was 66.7
^%
for Binary and 62% for Multiclass. On fusion, the highest accuracy obtained was 69.7% for Binary and 60.4% for Multiclass. Whereas in the second approach, the preprocessed images are converted into vector using the CV2 library in python. For the first two approaches the obtained feature files were implemented in all the four classifiers such as Support Vector Machine, Logistic Regression, Random Forest and XG Boost. In second approach, the highest performance in all parameters was obtained in XG Boost classifier. The highest accuracy obtained was the highest around 73% in Binary and 67% in Multiclass. In the Last approach, deep feature extraction was carried out and implemented in CNN model with SVM Classifier. The Accuracy obtained was maximum of about 75% for Binary which was the highest when compared with all the approaches. Also, other parameters significantly improved. The aim of this paper is to summarize all the three approaches and to provide the future scope.