ReVGG-R2Net: Optimized recurrent framework for microscopic blood cell segmentation.

Accurate segmentation of microscopic blood cell images is critical for advancing biomedical analysis and diagnostics. However, existing segmentation models often fall short due to limited dataset diversity and challenges in capturing fine cellular details, especially when different cell types are closely positioned. This paper addresses these challenges by introducing ReVGG-R2Net, a novel model architecture tailored for the precise segmentation of diverse blood cell types. ReVGG-R2Net integrates recurrent blocks within both the encoder and decoder, enhancing feature refinement to capture intricate cellular structures. The encoder employs a modified VGG16 backbone with recurrent features, facilitating effective spatial and contextual mapping, while the R2U-Net- based decoder leverages recurrent feature fusion to improve segmentation accuracy in regions with densely packed cells. Additionally, we present RaabinWBCSeg, a comprehensive dataset containing a broad spectrum of uninfected and infected cell types, filling gaps in existing benchmarks and promoting generalization in cell segmentation tasks. Extensive experiments were performed across five different benchmark datasets, including RaabinWBCSeg and BBBC041Seg, to thoroughly evaluate the proposed model's capability. ReVGG-R2Net consistently achieved state-of-the-art (SOTA) performance.