Our paper introduces an innovative method for image segmentation in total knee arthroplasty (TKA) using deep learning techniques with synthetic data. Traditional TKA image segmentation methods rely on costly X-ray images for training, posing accessibility and scalability challenges. In contrast, our method circumvents the need for expensive real X-ray data by utilizing synthetic images generated from 3-dimensional (3D) computer-aided-design (CAD) models. We generated synthetic data by projecting 20 CAD models of the knee joint onto images, creating a dataset of 10,000 images. These images were augmented with random X-ray knee backgrounds and Gaussian noise. This cost-effective and scalable approach trains deep neural networks (DNNs) in image segmentation tasks, ensuring robustness and generalization to real-world images. To assess efficacy, we conducted experiments using 4,781 X-ray images of TKA cases with known ground truth annotations. The trained neural networks were validated on this dataset, achieving a mean average precision (mAP) of 0.77. These results demonstrate our method's effectiveness in accurately detecting and segmenting the tibia and femur silhouettes in TKA images, eliminating the need for costly real X-ray data. In summary, our study highlights the potential of deep learning with synthetic data for medical image segmentation, particularly in TKA. By overcoming the limitations of real data acquisition, our approach enhances the accuracy and efficiency of image analysis in orthopedic surgery, advancing computer-assisted diagnosis and treatment planning.