Abstract: The early diagnosis of Alzheimer's disease (AD) through magnetic resonance imaging (MRI) analysis is of great significance. The detection of AD from neuroimaging data such as MRI through deep learning has become an attractive method. However, MRI image analysis based on deep learning often faces challenges such as small training sample size and high 3D data calculation costs. In order to get better model performance on small samples of MRI, this paper proposes a cascaded CNN and LSTM AD diagnosis model. Firstly, CNN is used to extract deep features of the image. Then, LSTM is used to classify sequence features. This treats 3D MRI images as sequences of 2D images, taking into account the correlation information between slice sequences. In addition, in order to improve the performance of the model on small sample data, knowledge distillation algorithms are used to obtain compressed lightweight model, and inter-teacher-student attention mechanism is introduced to improve the accuracy of model classification. The experiment shows that the diagnostic model can achieve better performance on the Alzheimer's disease neuroimaging initiative (ADNI) dataset.