The choice of loss function is crucial in medical image segmentation, as it directly influences the accuracy and robustness of the model. Conventional loss functions, such as Binary Cross-Entropy (BCE) and Dice loss, often suffer from imbalanced optimization, focusing too much on either region overlap or boundary accuracy, leading to suboptimal segmentation results. To address this limitation, we propose a dynamic weight-adjusted ensemble loss, called as DyWAEn loss, function that combines BCE, Dice, Hausdorff, and Tversky losses. Our method dynamically adjusts the weights of these loss functions based on their performance during training, allowing the model to emphasize the most relevant losses. This approach effectively mitigates the shortcomings of traditional loss functions by providing a balanced optimization across various metrics. Experimental results on the ultrasound and polyp image datasets demonstrate that the DyWAEn loss significantly outperforms individual loss functions in terms of Dice coefficient and Intersection-Over-Union (IoU). The DyWAEn loss function is highly adaptable and can be seamlessly incorporated into a wide range of segmentation frameworks without the need for adjustments, providing a flexible and effective solution for improving deep learning models in medical imaging.