PSPNet EPO-SEB: a novel attention-enhanced hybrid model for accurate histopathological image segmentation
Prem Purusottam Jena, Debahuti Mishra, Kaberi P. Das, Sashikala Mishra, Mandakini Priyadarshani Behera · Connection Science · 2025
Precise histopathological image segmentation is vital for accurate diagnosis and treatment planning. This manuscript proposes a hybrid framework, PSPNet EPO-SEB, combining PSPNet with an emperor penguin optimizer and an attention-enhanced module for improved segmentation performance. The model was rigorously evaluated on two prominent datasets, BACH and Camelyon17, encompassing high-resolution and whole-slide histopathological images, respectively. Experimental results demonstrate that PSPNet EPO-SEB outperforms conventional segmentation models, achieving dice coefficients (DC) of 0.9237 and 0.9186, and intersection over union (IoU) values of 0.8629 and 0.8622 on the BACH and Camelyon17 datasets, respectively. These metrics surpass those of competing models such as U-Net, V-Net, PA-Net, FANet18, Mask R-CNN, R2UNet, with PSPNet EPO-SEB showing enhanced boundary accuracy, True positive rates (TPR) above 0.93, and minimized false positive rates (FPR) at 0.1211 on BACH and 0.1108 on Camelyon17. Furthermore, the proposed model maintains low average error rates (AER) and achieves boundary precision with Hausdorff distances (HD) as low as 12.68 on BACH and 13.04 on Camelyon17, underscoring its accuracy in delineating complex tissue structures. Despite a slight increase in computational time due to optimization and attention mechanisms, the enhanced segmentation precision and boundary adherence make PSPNet EPO-SEB a highly effective solution for complex histopathological image analysis.