Effective classification of brain MRI images is critical for diagnosing and managing brain tumors, one of the most prevalent forms of cancer. The complexity of these images and the need for precise categorization drive the demand for advanced analytical methods. Long Short-Term Memory (LSTM) networks, known for their capacity to model sequential data and capture intricate patterns, offer a powerful approach for analyzing such image data. Optimizing LSTM networks, however, poses challenges due to the sensitivity of their performance to hyperparameter settings. Traditional methods for tuning these parameters often rely on gradient-based optimization techniques, which may fall short in exploring the complex and multidimensional parameter space. In this study, we introduce an innovative approach by integrating Particle Swarm Optimization (PSO) with LSTM networks to enhance the classification of brain MRI images. PSO, an optimization technique inspired by natural swarming behavior, provides a robust mechanism for searching the hyperparameter space more efficiently than conventional gradient-based methods. By applying PSO, we aim to fine-tune the LSTM network’s parameters such as the number of units, learning rates, and dropout rates thereby improving its classification accuracy.