A Brief Experimental Survey of Semantic Image Segmentation on PanNuke and NuCLS Datasets

Semantic image segmentation is changing digital pathology by making it faster and more accurate to analyze tissue samples through digital images. Whole Slide Imaging converts these samples into high-resolution digital formats, allowing for detailed analysis. Traditional semantic segmentation uses deep learning models, but evaluations have mainly focused on specific datasets, limiting broader applicability. This paper presents a brief experimental survey of semantic image segmentation on two well-known datasets: PanNuke and NuCLS. We assess the performance of traditional deep learning models on these datasets, highlighting recent trends and challenges. Our experimental survey aims to help the understanding of semantic image segmentation and eventually to the field of digital pathology advance, making diagnosis more accurate and accessible. Our findings can guide researchers and clinicians in selecting appropriate models for their specific needs, ultimately improving patient care through more precise disease detection.