Recent advances in deep learning have promoted the development of many skin lesion segmentation models; however, traditional architectures still do not fully exploit saliency information in dermatological images. In this study, we introduce SAL-GS-TransUNet, an improved variant of GS-TransUNet that integrates 2D Gaussian Splatting with a multiscale saliency-fusion branch to enhance feature representation and refine lesion boundaries. The model architecture consists of a GSConvBlock encoder, a Transformer bridge for global context learning, a decoder with skip connections, and an independent saliency branch based on GaussianSplat2D fused into the final output. The model was evaluated on the ISIC-2017 and PH2 datasets using 5-fold cross-validation, achieving stable performance with F1 (Dice) 0.878 ± 0.0218, AUC 0.982 ± 0.0086, Accuracy 0.950 ± 0.0094, along with a balanced trade-off between Precision and Recall. These results demonstrate that the saliency branch improves both generalization and training stability. This highlights the potential of SAL-GS-TransUNet for automated diagnostic systems and future multitask medical image analysis.