Semantic segmentation of medical images is essential for enhancing computer-aided diagnosis and treatment planning. This work addresses the multi-category semantic segmentation of CT images, targeting five categories: Background, Liver, Kidney, Spleen, and Pancreas. Two state-of-the-art models, UNet++ and DeepLabv3, pretrained on the ImageNet dataset, are employed to benchmark performance. Additionally, a custom-designed model is implemented, incorporating a UNet backbone with residual blocks in the encoder and advanced attention mechanisms such as attention gates and squeeze-and-excitation blocks in the decoder. The models are evaluated using precision, recall, F1-score, IoU score, and Dice score metrics for each organ. Comparative analysis reveals the strengths and limitations of each approach, with the custom model showcasing significant improvements in capturing finer anatomical structures. The results contribute to the ongoing advancement of deep learning-based segmentation techniques for medical imaging.