Identification of Alzheimer’s Disease using Deep Learning with the multi-layer perceptron (MLP) Classifier

Abstract

Throughout the past decade, the world cautiously predicts the potential impact of  artificial intelligence (AI) on daily life from education to healthcare. This study examined the accuracy and efficiency of a deep learning model when detecting Alzheimer's disease in brain MRIs. Python code was used to create a model that would compute the accuracy and error loss with different hyper parameters. The multi-layer perceptron (MLP) Classifier was used to train the model and the classifier includes multiple hyper parameters that can be fine-tuned to improve the accuracy of the model. Next, we focused on optimizing the values of two hyper parameters: learning rate and hidden layer sizes. The model performed best with a learning rate of 0.0001 and an accuracy of 97.19%. For hidden layer sizes, the number of neurons per layer was optimized to compute the highest accuracy. The model performed best with one layer of 50 neurons with an accuracy of 95.94%. When both hyper parameters were changed in the same experiment, the accuracy decreased to 96.09%. The optimum model with a hidden layer size of two layers of 50 neurons and a learning rate of 0.0001 earned a sensitivity of 95.50% and a specificity of 99.02%. We observed that an accuracy above 90% can only be reached by optimizing with each hyper parameter. This study demonstrates the feasibility of using deep learning modeling with the MLP Classifier for successful identification of Alzheimer’s disease.