Impact of Image Resolution on Convolutional Neural Networks for Alzheimer Disease Dataset Classification: A ResNet Study

Reducing image resolution can save computing resources but may result in loss of details. Understanding the relationship between image resolution and the performance of various CNN architectures is crucial for setting standards in CNN -based model properties. This study investigates CNN performance on the Kaggle Alzheimer's dataset, comprising 6,400 images divided into four sections representing disease progression stages. We evaluated the ResNet CNN variants models for image classification using resolutions of 64x64, 128x128, and 256x256 pixels. Performance was assessed using evaluation metrics including Accuracy, Precision, Recall, F -Measure, Loss and Time of execution. Optimal performance, with the highest evaluation metric values, was observed with the highest resolution (256x256 pixels), achieving classification of the entire dataset, including all subclasses. This highlights the significant influence of image resolution on CNN models performance. Moreover, resolution affects the computational time and complexity of calculations. As an example of the results, among the best performances obtained, the ResNet-101 model with a resolution of 256x256 achieved an accuracy of 97.08%, a precision of 97.76%, a recall of 98.13%, and an F -measure of 97.94<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, with a loss of 0.045. However, the average execution time per epoch is about 75 seconds, which is a significant time.