Development of a Method for the Early Detection of Alzheimer Using CT Images, Deep Learning Techniques and Hyper-parameter Tuning

The world health organization (WHO) has published an article that raises public awareness about dementia, from which they declared that in 2019 around 35.6 million cases were diagnosed. Experts predict that this number will double by the year 2030 and triple by the year 2050. Dementia can be caused by a variety of factors, but the biggest cause of dementia is Alzheimer’s disease. For this reason, this research is focused on the construction of an architecture of a model that classifies images. The architecture for the methodology is structured in 5 phases. The first phase is the recollection of data from a public website like Kaggle that handles more than 40K images from which each image is classified in 4 categories (non demented, mild demented, moderate demented, and very mild demented). The second phase is the preparation of the images, and we do so by resizing the images and applying noise reduction filters if necessary. In the third phase we train the Convolutional Neural Network (CNN) with the data that we have prepared, as an additional boost to this phase, hyperparameter tuning is applied to obtain the best parameters thus obtaining the best results. In the fourth phase we test the model and evaluate its classification using quality measurements. In the final phase we interpret the results obtained by the model in conclusion the model obtained an accuracy of 97.39%. As a proposal for future work, we propose the implementation if an API to increase the dataset and obtain new information resulting in an interesting analysis of the behavior of the model as the dataset increases.