Ultrasound imaging technology plays a vital role in medical imaging. Ovarian ultrasound image segmentation is challenging due to the wide variation in lesion sizes caused by the cancer detection period and individual differences, as well as the noise from reflected wave interference. To address these challenges, we propose an innovative algorithm for ovarian ultrasound image segmentation that incorporates multi-scale features. This algorithm effectively processes image data with varying scales. By introducing a skip connection structure, the shallow image features are preserved. Additionally, in the feature fusion module, feature maps extracted from the backbone network are integrated layer by layer, enhancing the model's ability to parse multi-scale features. The proposed algorithm was tested on ovarian ultrasound images that had undergone noise reduction using different filtering methods. When compared to mainstream segmentation algorithms, our model achieved improvements in mIoU, mAcc, and aAcc metrics by 2.02, 1.09, and 0.34%, respectively. Overall, the algorithm outperformed the comparison methods, offering a new solution for ovarian ultrasound image segmentation.