Accurate segmentation of papillary thyroid microcarcinomas (PTMCs) during ultrasound-guided radiofrequency ablation (RFA) is critical for effective treatment, but is challenging due to acoustic artifacts, small lesion size, and anatomical variations. In this study, we propose DualSwinUnet++, a dual-decoder transformer-based architecture designed to improve PTMC segmentation by incorporating thyroid gland context. DualSwinUnet++ uses independent linear projection heads for each decoder and a residual information flow mechanism that transfers intermediate features from the first (thyroid) decoder to the second (PTMC) decoder via concatenation and transformation. These design choices allow the model to explicitly condition tumor prediction on thyroid morphology without shared gradient interference. Trained on a clinical ultrasound dataset with 691 annotated RFA images and evaluated against state-of-the-art models, DualSwinUnet++ achieves excellent Dice and Jaccard scores while maintaining an inference latency of less than 200 ms. The results demonstrate that the model is suitable for near-real-time surgical assistance and is effective in improving segmentation accuracy in challenging PTMC cases.
