LAUSR

This paper proposes a multitask deep learning framework for simultaneous skin lesion segmentation and classification, tailored for resource‐constrained environments. The architecture integrates ShuffleNet as an encoder within a U‐Net framework as a decoder to extract features. Using ShuffleNet's depthwise separable convolutions and channel shuffling, the model captures both low‐level spatial details and high‐level semantic features. To enhance feature representations, the proposed adaptive multiscale dynamic attention network (AMSDAN) is employed at two critical stages: first, to refine feature maps during segmentation by dynamically normalizing features across multiple scales; and second, to refine the segmented output for classification by highlighting discriminative patterns. The proposed work is trained and validated using HAM10000 and ISIC‐2017 datasets. Experimental results demonstrate that the framework outperforms state‐of‐the‐art techniques.