Abstract
Computational pathology datasets are commonly described using inconsistent and locally defined terminology, often conflating biological units (patients, specimens), lab preparations (blocks, slides), and derived digital data (whole-slide images (WSIs), tiles, and patches). This ambiguity complicates interpretation of training data composition, limits comparability across studies, and affects claims regarding dataset scale, generalizability, and clinical relevance. We reviewed widely cited foundation models and observed substantial variability in how disease representation is reported, highlighting the need for a standardized hierarchical framework. We propose a straightforward harmonization framework by aligning dataset descriptions with the hierarchical information model defined by DICOM. The framework distinguishes clinical, lab, and digital domains and recommends explicit reporting across hierarchical levels. Adoption of this terminology enables transparent dataset characterization, improves reproducibility, and facilitates regulatory and clinical translation of computational pathology. This pragmatic framework provides an immediately implementable pathway towards harmonized reporting in artificial intelligence-driven pathology research.