In this chapter, we propose the global–local attention block (GLA-Block), a novel attention module that flexibly integrates multi-head self-attention with lightweight attention mechanisms (CBAM, BAM, and scSE) to simultaneously capture global context and fine-grained local details in CNNs. The GLA-Block is inserted after the third stage of two popular backbones, VGG16 and ResNet18, enabling concurrent channel–spatial attention with multiple heads. We evaluate our method on two dermoscopic image classification datasets, ISIC-2018 and HAM10000. Experimental results demonstrate that, compared to the baseline and single-module attention variants, the GLA-Block improves Top-1 accuracy by approximately 1–2% points on ISIC-2018 and by approximately 6–8% points on HAM10000, while increasing the parameter count by less than 2% relative to the original backbones. These gains hold in both ImageNet-1K pretrained and training-from-scratch settings. The main contributions of this work include: (1) the design of the GLA-Block for effective fusion of self-attention and lightweight attention modules; (2) a rigorous evaluation on two biomedical datasets to verify consistent performance gains; and (3) an in-depth analysis of the trade-off between accuracy and model complexity. Our findings show that the GLA-Block is an efficient and parameter-economical solution, well-suited for dermoscopic image classification applications.