Self-Supervised Denoising of Thyroid Ultrasound Images Using SE-Module Enhanced U-Net with FPN

作者：Changhao Sun, Jianning Chi, Haojia Yu, Bo Wu, Zelan Li, Ying Huang · 年份：2025 · DOI：10.1109/ccdc65474.2025.11090577 · 被引用次数：2 · 研究领域：Image and Signal Denoising Methods

Thyroid ultrasound imaging plays a crucial role in the diagnosis of thyroid diseases, but the quality of ultrasound images is often compromised by noise, leading to suboptimal image quality. Additionally, collecting paired clean and noisy images from real-world scenarios is extremely costly and timeconsuming. To address these challenges, we propose a selfsupervised denoising method that integrates an SE (Squeezeand-Excitation) module into the last layer of the U-Net encoder to enhance feature map channel dependencies, and combines it with an FPN (Feature Pyramid Network) module to handle multiscale targets. This approach leverages deep learning to improve the quality of thyroid ultrasound images without the need for paired clean and noisy data. Experimental results on our thyroid ultrasound image dataset demonstrate that our proposed method achieves a PSNR improvement of 0.8 to 0.9 dB. Future work will focus on refining and extending this model to other medical image sequences, including CT and MRI slices.