Sickle Cell Disease (SCD) is a life-threatening genetic condition that causes red blood cells to become sickleshaped, leading to serious health complications. Current diagnostic methods rely on manual inspection of microscopic blood images, which can be slow, subjective, and error-prone. This study presents a comparative analysis of deep learning architectures to automate SCD diagnosis. Using a publicly available image dataset, we applied preprocessing techniques including data augmentation, class balancing, and PCA visualization. We systematically trained and evaluated both Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) models to classify red blood cells as healthy or sickle-affected. Our comparative evaluation revealed significant performance differences between the architectures. The CNN model achieved superior results with 100 % training accuracy, 96.12 % validation accuracy, 93.85 % test accuracy, and an AUC score of 0.94, while the LSTM model demonstrated lower performance across all metrics. The CNN also showed better balanced performance across both classes, effectively addressing class imbalance challenges. This comparative study demonstrates that CNN-based approaches provide faster, more consistent, and scalable solutions for SCD detection compared to LSTM models, particularly valuable in resource-limited healthcare settings. By optimizing architecture selection through systematic comparison, this work supports automated early detection and reduces dependency on manual diagnosis, ultimately bringing reliable SCD screening closer to patients who need it most.