Tran et al. adopts a human-in-the-loop semi-supervised framework for cervical ultrasound image segmentation based on a U-Net architecture. The model is initially trained on a small labeled set of 50 images using a combination of Dice and cross-entropy (CE) losses. Pseudo-labels for the unlabeled data are then generated and iteratively refined by expert annotators using Label Studio, with particular emphasis on correcting major segmentation errors such as disconnected regions and inaccurate boundaries. These refined masks are progressively incorporated into the training set over multiple iterations, resulting in continuous performance improvement. In addition, all source code used for pseudo-label generation, together with the pseudo-labels produced at each refinement stage, has been publicly released to support reproducibility and further research. Tran, Nam-Khanh, et al. "Human-in-the-Loop Semi-Supervised Uterine Cervix Ultrasound Image Segmentation." 2025 IEEE 22nd International Symposium on Biomedical Imaging (ISBI). IEEE, 2025. Jieyun Bai et al. "FUGC: Benchmarking Semi-Supervised Learning Methods for Cervical Segmentation" 2026 IEEE TMI