The SALV-Dataset Registry: An Expertly Curated Digital Clinicopathological Dataset for Salivary Gland Tumor Research and AI-Assisted Diagnostic Tools.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
UNLABELLED: Salivary gland tumors are rare and morphologically diverse, posing both diagnostic and scientific challenges. This study presents the first phase of the SALV-Dataset Registry; a nationwide, expertly curated, and fully digitized clinicopathological resource, designed to support research and develop artificial intelligence (AI) tools assisting in salivary gland tumor pathology diagnostics. METHODS: Salivary gland tumor resections diagnosed at the Leiden University Medical Center (1999-2024) were collected through the Dutch national network and registry for histo- and cytopathology (PALGA). In total, 685 cases were included. Hematoxylin- and eosin-stained slides were digitized and independently reviewed by three teams of head and neck pathologists, in line with the 2023 WHO Classification of Head and Neck Tumours. Discordant and ambiguous cases were resolved in consensus meetings, with access to immunohistochemistry, molecular analysis, and clinical data. Interobserver agreement among the three teams was quantified (Fleiss' kappa), and agreement between the original and consensus diagnosis was determined (Cohen's kappa). RESULTS: Of the 685 tumors, 75% were benign, 24% malignant, and 1% of uncertain malignant potential. The parotid gland was most frequently involved (86%), and the highest rate of malignancy was observed in the sublingual gland (100%). Rare entities were represented, although numbers remained limited. Interobserver agreement was substantial (κ of 0.64; 95% 0.60-0.67). Agreement between the original and consensus diagnosis was excellent (Cohen's κ of 0.89; 95% CI 0.86-0.92). Forty-three cases (6%) were reclassified following revision, with a change in diagnostic category in 11 cases (2%). CONCLUSION: This study demonstrates the feasibility of large-scale, expert digital revision of salivary gland tumors, and establishes a robust foundation for future clinicopathological and AI-based research. With national expansion ongoing, the SALV-Dataset Registry will provide a comprehensive resource for AI training, validation, and clinically oriented modeling in salivary gland tumor diagnostics.