Cleft lip, a common congenital facial deformity, occurs during early pregnancy due to incomplete oral and labial fusion, affecting facial symmetry and aesthetics. Despite advanced medical imaging models for automated detection and classification, the disease remains scarce. This study proposes a novel deep learning (DL) model using YOLOv8 to detect and classify cleft lip, a common congenital facial deformity in early pregnancy. The model uses a features extraction method and spatial attention mechanism to extract discriminative features from facial images, ensuring high accuracy and reliability in detecting cleft-related anomalies. The model achieves classification accuracy of 97.3%, precision of 96.1%, recall of 97.7%, and a mAP50-95 of 74.8%, demonstrating its effectiveness in both diagnosis and post-surgical assessments of facial asymmetry. However, the study acknowledges its limitations, such as dataset size and generalizability. Future research should include broader dataset inclusion and real-time application development. The study emphasizes the applicability of integrating DL models into medical diagnosis systems to assist healthcare professionals and improve patient outcomes.