The irreversible nature of Alzheimer’s disease (AD) places a heavy burden on patients and their families. Mild Cognitive Impairment (MCI) is a transitional stage between normal aging and AD, and its early detection is essential for timely intervention. With the advent of deep learning, convolutional neural networks have shown promise in diagnosing AD using MRI scans. This paper introduces an improved image classification model based on ResNet-50 and the Contextual Transformer (CoT) block, which combines global and local information capture to enhance feature representation. The experimental results demonstrated that the model achieved high accuracy in classifying AD, MCI, and healthy controls (HC), suggesting that deep learning holds great potential for improving AD diagnosis.