Accurate classification of brain infarcts in CT scans is important for timely medical intervention. However, deep learning models often struggle with low-quality, unstructured datasets. This study enhances the performance of the ResNet18 Convolutional Neural Network (CNN) by integrating refined methods for data cleansing and transformation to address these challenges. Using a publicly available Kaggle dataset, we mitigated issues of noise and disorganization through meticulous data cleaning and restructuring, ensuring an optimized training set. The refined ResNet18 model effectively classified CT scans into stroke and non-stroke categories, achieving notable improvements in accuracy. Our findings emphasize the critical role of high-quality data preparation and model optimization in medical imaging, offering a practical and replicable approach for to advancing brain infarct detection in clinical and research settings.