Intelligent diagnosis for endometrial hyperplasia and endometrial cancer: a comprehensive survey and taxonomy

Endometrial hyperplasia (EH) and endometrial cancer (EC) are common gynecological conditions. This study aimed to survey intelligent diagnosis research for EH and EC, with a focus on a standardized taxonomy of diagnostic features, systematic dataset assessment, and the capabilities of diagnostic tools. We conducted a systematic literature review following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, searching across Web of Science, Scopus, PubMed, IEEE Xplore, ACM Digital Library, DBLP, and Google Scholar on August 8, 2025. We additionally searched grey literature on GitHub on September 11, 2025. A total of 1,096 records were initially retrieved from multiple databases. After removing duplicates, 516 unique records entered title and abstract screening. After full-text assessment, 54 scientific articles and 9 grey-literature tools were included in this study. A comprehensive taxonomy of EH and EC diagnostic features was developed based on the diagnostic features mentioned in the studies we retrieved and the guidance provided by two gynecologists. Datasets were analyzed across dimensions of input type, scale, openness, license, usage frequency, and feature coverage. The analysis of EH and EC intelligent diagnostic tools includes algorithm type, technique, input data type, performance evaluation indicators, clinical validation status, and availability. After screening 516 deduplicated articles, 54 scientific articles were included. In addition, 9 grey-literature tools were identified through grey literature retrieval. Based on 54 scientific articles and 9 grey-literature tools, we constructed a comprehensive taxonomy encompassing 313 distinct diagnostic features organized into 9 categories and 35 subcategories, with all features annotated with acquisition methods and difficulty levels. Analysis of 54 datasets revealed significant limitations: poor openness (only 7 being publicly available), dominance of unimodal image data (75.9% of datasets), and inadequate feature coverage (43.5%). Evaluation of the 63 intelligent diagnostic tools showed that most lack rigorous clinical validation and multimodal data integration capability, substantially limiting their clinical applicability. The proposed taxonomy offers a standardized framework for feature selection, model comparison, and tool evaluation in EH and EC intelligent diagnosis. These findings highlight critical gaps in current datasets and tools, providing clear directions for developing more robust, interpretable, and clinically applicable intelligent diagnostic tools for EH and EC.