Alzheimer's Disease Classification Using Convolutional Neural Network and Random Hyper-Parameter Tuning

The most prevalent kind of dementia is Alzheimer's disease (AD). Damage or death to nerve cells and their connections in the brain causes it. This study conducted a thorough examination of the available neuroimaging data to identify AD in a timely manner. In this study, we present a comprehensive investigation into Alzheimer's stage classification and prediction using the ADNI MRI dataset with a convolutional neural network. We resized the dataset to a resolution of (100,100) and converted it to grayscale. We preprocessed the dataset using zooming and recycling, treating flips as false due to the significance of orientation in MRI datasets. We divided the dataset into an 8:1:1 ratio for training, testing, and validation, respectively. Then, we put the data into a CNN model and added different convolutional layers, batch normalization layers, dropouts to deal with overfitting, the activation function, and leaky Relu. We assembled the model using the Adam optimizer and a loss function based on categorical cross-entropy. Next, we looped the primary model through the Keras random hyperparameter tuner, executing five trials while considering the validation accuracy. After tuning the hyperparameters, we obtain our best model. Our research revealed that the accuracy for different stages of Alzheimer's disease was 97% for CN, 96.95% for EMCI, 99.99% for LMCI, 99.23% for MCI, and 99.99% for AD. Our research yielded an overall accuracy of 97.86% for the best model, with an Fl_score ranging from 0.98 to 1.00. We evaluated the model using a confusion matrix and a classification report, finding it superior to other models created with the ADNI dataset and CNN.