Dementia is a cognitive decline of the brain’s functionalities that impacts the individual’s memory, reasoning, and communication skills. Many neurodegenerative disorders like Alzheimer’s disease or Lewy body dementia are the primary cause of dementia. Practitioners mostly employ imaging modalities including CT scans, MRI scans, and PET scans to diagnose dementia. Deep learning architectures were embraced by the research community to automate the screening process and increase productivity while also boosting the diagnosis performance. This work designs a convolutional network based on squeeze and excitation blocks (SE Network) for predicting the dementia grades; moreover, other deep models (such as baseline CNNs, pretrained networks, and vision transformers) are also adapted to the issue of dementia classification. The conducted experiments showed that the baseline CNN, SE Network, and a majority vote of SE networks achieved a test accuracy of 58.52%, 70.73%, and 74.43%, respectively. Additionally, the squeeze and excitation networks demonstrated a notable performance with respect to other available state-of-the-art CNN networks.