Human Organ Functional Tissue Units Image Segmentation Based on UNeXt50

The automatic mask segmentation of Functional Tissue Units (FTU) in human organs is a critical task for understanding cellular/tissue interactions and their impacts on human health. However, this task is challenged by the need for cross-organ generalization, ambiguous FTU boundaries, highresolution image complexity, and cross-dataset domain adaptation. To address these issues, we propose a UNeXt50based deep learning framework, which integrates ResNeXt50, Atrous Spatial Pyramid Pooling (ASPP), and Feature Pyramid Network (FPN) to enhance multi-scale feature extraction and context aggregation. Using a hybrid loss function combining Dice loss and cross-entropy loss, the model achieves robust performance across five human organs (prostate, spleen, lung, kidney, large intestine). Experimental results on the HuBMAP + HPA dataset demonstrate that UNeXt50 achieves an optimal Dice coefficient of 83.7% with a threshold of 0.51, outperforming traditional U-Net variants in handling complex tissue boundaries and cross-organ variability. This study provides a scalable solution for large-scale FTU segmentation, facilitating advancements in cellular biology and precision medicine.