Abstract <p>Malaria is a severe infectious disease caused by <i>Plasmodium</i>, posing a significant threat to public health. Accurate detection of <i>Plasmodium</i> in blood cell images is critical for timely diagnosis and treatment. However, existing Transformer-based models face limitations in processing the inherent scale and morphological diversity of infected cells, particularly in multi-scale perception and fine-grained feature extraction. To address these challenges, we propose a novel deep learning architecture for malaria cell classification, named ISTNet (Inception Swin Transformer Network). Specifically, it introduces a continuous ISformer module to integrate the multi-scale feature extraction capability of the Inception structure with the global attention mechanism of the Swin Transformer. Furthermore, a multi-branch Inception Mixer block is proposed to enhance receptive field diversity and improve the model’s ability to distinguish cellular infection features. ISTNet’s hierarchical feature fusion architecture effectively overcomes scale sensitivity and detail degradation in <i>Plasmodium</i> recognition. Comprehensive experiments on two public datasets are conducted to validate the effectiveness of the proposed method. Our model achieves classification accuracies of 96.61% and 99.65% on the NIH and BBBC041 datasets, respectively. This work provides a robust solution for intelligent malaria diagnosis, with practical value for primary healthcare in resource-limited settings. It also demonstrates strong generalizability to other medical image analysis tasks.</p> Graphical abstract <p></p>