BMC research notes · 2026
A domain-specific annotation framework and expert-annotated corpus for structured analysis of German HNSCC pathology reports.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Gubik Sebastian, Seger Franziska, Vollmer Andreas, Hörner Marius, Renner Tobias, Kohlhepp Luca, Manthey Marc, Müller-Richter Urs, Kübler Alexander, Hackenberg Stephan, Hartmann Stefan
Аннотация
OBJECTIVE: To develop and empirically characterize a domain-specific, block-structured annotation schema and an expert-annotated reference corpus for German-language head and neck squamous cell carcinoma (HNSCC) pathology reports.
RESULTS DESCRIPTION: We retrieved 631 pathology reports from 145 patients treated at University Hospital Würzburg between January 2021 and March 2024 and split them at the patient level into training, validation, and evaluation subsets. Reports were manually annotated using a hierarchical, span-based schema supporting overlapping and nested annotations. The resulting corpus contains 45 unique labels organized into clinically meaningful blocks and an average of 37-39 annotations per report across subsets. Label frequencies and normalized category distributions were broadly comparable across the three subsets. Fine-grained labels were predominantly applied within their designated blocks, indicating adherence to the intended block structure. The corpus provides a domain-specific reference resource for developing and evaluating structured information-extraction approaches for German HNSCC pathology reports.
RESULTS DESCRIPTION: We retrieved 631 pathology reports from 145 patients treated at University Hospital Würzburg between January 2021 and March 2024 and split them at the patient level into training, validation, and evaluation subsets. Reports were manually annotated using a hierarchical, span-based schema supporting overlapping and nested annotations. The resulting corpus contains 45 unique labels organized into clinically meaningful blocks and an average of 37-39 annotations per report across subsets. Label frequencies and normalized category distributions were broadly comparable across the three subsets. Fine-grained labels were predominantly applied within their designated blocks, indicating adherence to the intended block structure. The corpus provides a domain-specific reference resource for developing and evaluating structured information-extraction approaches for German HNSCC pathology reports.