<p>Computer-aided liver tumor segmentation plays a crucial role in reducing healthcare professionals’ workload and improving surgical outcomes. Traditional segmentation methods offer limited automation, while deep learning approaches can suffer from overfitting due to insufficient data, reduced accuracy, and high hardware demands. This paper proposes combining an improved Yolo-v5 method with the B-spline level set method to address these issues. The Yolo-v5 model identifies and localizes liver tumors, providing an initial contour for the level set method and enhancing the automation of traditional approaches. Notably, Yolo-v5 operates efficiently on both GPU and CPU environments, without being limited by hardware. Our improvements to Yolo-v5 include optimizing the activation function, refining the loss function, and developing a lightweight neural network model. These updates enable Yolo-v5 to generate a more accurate initial contour for the B-spline level set method, which is further refined using a Gaussian mixed model to improve segmentation accuracy. Experimental results demonstrate that our enhanced Yolo-v5 model outperforms 3DUnet on the LiTS dataset, achieving a 3.9% increase in mean Average Precision at IoU 0.5 (mAP50) and a 10% improvement in mean Average Precision across multiple IoU thresholds (mAP50:95). Furthermore, our method yields a Dice coefficient that is 2.77% higher than 3DUnet’s. In real-world tests, the combined Yolo-v5 and B-spline level set approach shows a 2.58% improvement over 3DUnet and a 3.3% enhancement compared to conventional factorization methods. These results demonstrate the potential of our algorithm to improve automatic liver tumor segmentation, leading to better clinical outcomes and more efficient medical imaging.</p>