As part of the George B. Moody PhysioNet Challenge 2026, we developed a hybrid Vision Transformer (ViT) and Long Short-Term Memory (LSTM) deep learning (DL) algorithm to predict cognitive decline with polysomnography (PSG) data, under the team name CorazonCansado. Studies are finding that sleep disturbances are increasingly associated with cognitive decline, supporting the use of multimodal PSG signals for predictive modeling as an early, non-invasive detection method. Using multimodal PSG data, including EEG, ECG, EOG, EMG, and respiratory signals, we developed a unified DL framework. The ViT component identifies complex patterns across multimodal signals, while LSTM captures how these patterns change over time. Using data from the Human Sleep Project provided by the challenge, we developed a predictive model to determine the likelihood of developing cognitive impairment later in life. The model was trained using balanced data to reduce potential bias and improve generalizability across diverse populations. Given the correlations between sleep quality and cognitive disease, the PSG data produced in sleep studies may prove a critical resource for predicting future cognitive decline.