This research investigates the feasibility of a communication system using Electroencephalography (EEG) signals mediated by a deep learning model, designed to aid individuals with significant communication challenges. The study builds on prior work in EEG-based image generation, testing the accuracy and efficiency of the system in conveying intended meanings. In the experiment, participants were presented images generated from EEG signals and were asked to give titles according to their interpretation. These titles were analyzed using text embeddings derived from a large language model (LLM) to measure cosine similarity. Results indicate that while sender and receiver interpretations often diverged, consistent patterns emerged within and between receivers. This suggests that repeated communication trials will align interpretations over time, improving mutual understanding. The findings highlight the potential of the method to facilitate adaptive communication, though further research is required to optimize its reliability, scalability, and practical applicability.