Recently, detecting malignant cells in leukemia cases has been rising as one of many challenging topics in applying deep learning in medical imaging to assist diagnostics. The detection of leukemia necessitates deep learning architectures that can distinguish between malignant and healthy cells, as their patterns often resemble each other on peripheral blood smear images. Therefore, this study proposes the combination of the Global Context and Local Interaction (GCLI) module to be fused in the YOLOv11 architecture, resulting in a replacement of the self-attention mechanism–the C2PSA block in the baseline version. The GCLI module helps the deep learning architecture focus on important features of targeting objects while suppressing the noise of background and other similar characteristic patterns by introducing the long-range dependencies capturing and the interaction between global and local features through 1D convolution to avoid dimensionality reduction during the attention weights generation process. As a result, the YOLOv11 with the GCLI module outperforms the baseline model and most of the other attention-based YOLOv11 models in high-precision leukemia detection, reaching up to 25.1% mAP50-95 in the LeukemiaAttri dataset while consuming less than 11.0% in parameters compared to the original YOLOv11. The study suggests that an attention module could be useful for assisting the deep learning models of detecting abnormal cells on peripheral blood smear images. On the other hand, GCLI indicates that the non-local interaction between pixels could provide more useful feature visualization for leukemia detection, which would improve the overall performance of CAD systems in medical imaging.