Detection of Alzheimer's Disease Using Image Processing and Deep Learning Methods: A Review

Alzheimer's disease (AD) is one of the most prevalent neurodegenerative diseases today, and early diagnosis directly affects the course of the disease. The aim of this study is to comparatively evaluate the national and international literature on the detection of AD using image processing and deep learning methods within a single framework. Following the PRISMA flow logic, a total of 28 original articles published between 2018 and 2025 were reviewed: 16 from TRDizin and 12 from international databases (PubMed, Scopus, Web of Science). Of the reviewed studies, 23 are based on magnetic resonance (MR) images and 5 on electroencephalography (EEG) signals. For each study, the dataset, sample size, architecture or algorithm, preprocessing steps and reported accuracy were extracted. The findings show that convolutional neural network (CNN) based methods and transfer learning are the dominant approaches on MR images, while the Kaggle 4-class dataset and ADNI are the most frequently preferred data sources. Architectures such as AlexNet, VGG-16, ResNet, DenseNet, MobileNetV2 and ConvNext yielded accuracies between 88% and 100%. On EEG signals, the highest reported accuracy was 96.59% using support vector machine (SVM) and k-nearest neighbour (kNN) methods. Studies conducted in Türkiye are competitive with the international literature in terms of accuracy; however, 11 of the 16 Turkish studies reviewed used the same open-access dataset and none reported clinical validation. Therefore, dataset diversity, external validity and clinical validation stand out as the primary shortcomings of the field.