What question did this study set out to answer?

The research aims to enhance the diagnosis of Alzheimer's disease utilizing AI techniques, specifically a deep super-resolution GAN framework.

May 9, 2026Open Access

A deep super-resolution GAN framework integrating ResNet-18 and CNN for enhanced Alzheimer’s disease diagnosis

Key Points

The research aims to enhance the diagnosis of Alzheimer's disease utilizing AI techniques, specifically a deep super-resolution GAN framework.
Implemented a deep super-resolution GAN with ResNet-18 and CNN models for image classification.
Utilized a dataset of 6400 MRI images for training and evaluation.
Employed hyperparameter tuning and data augmentation to optimize model performance.
ResNet-18 achieved a test accuracy of 97.50% and an AUC of 100%.
CNN model reached a test accuracy of 99.22% and an AUC of 99.8%.
DSR-GAN showed an SSIM level of 0.847 and PSNR of 29.30 dB.

Abstract

Medical imaging techniques, such as computerized tomography (CT), magnetic resonance imaging (MRI), and ultrasound, provide critical insights for precise disease diagnosis and detection. Despite their importance, these imaging methods are often expensive to produce and store due to their substantial data size and complexity. Moreover, interpreting these images manually can be a time-intensive task, further compounded by the rising demands on healthcare systems worldwide. Artificial intelligence (AI) techniques offer a promising solution to address these challenges. For instance, deep learning (DL) can be utilized for tasks such as medical image classification. This paper proposes two approaches: a deep super-resolution generative adversarial network (DSR-GAN) with a ResNet-18 model and a DSR-GAN with a convolutional neural network (CNN) model to recognize four types of Alzheimer's disease (AD): Mild-Demented (MD), Moderate-Demented (MOD), Non-Demented (ND), and Very-Mild-Demented (VMD). The DSR-GAN is applied via the PyTorch framework with a dataset comprising 6400 MRI images. A super-resolution (SR) approach is employed to improve the images' clarity, enabling the proposed DSR-GAN to enhance specific aspects of images. The ResNet-18 and CNN models undergo hyperparameter tuning, and data augmentation techniques are utilized to optimize efficiency. Error matrix is experimentally utilized to measure the efficiency of the ResNet-18 and CNN. The ResNet-18 model reached a test accuracy (TA) of 97.50% and an area under the curve (AUC) of 100%, while the CNN model reached a TA of 99.22% and an AUC of 99.8%. Structural-similarity-index-measure (SSIM) and peak-signal-to-noise ratio (PSNR) measurements are employed to assess the efficiency of the DSR-GAN. It achieved an SSIM level of 0.847 and a PSNR of 29.30 dB. The mixture of DSR-GAN with ResNet-18 and CNN models offers a fast and accurate method of differentiating between Alzheimer's cases and could assist professionals in screening these cases with AD.

اسأل الذكاء الاصطناعي

Bookmark

View Full Paper