Service Specific Micro Adaptation for Specimen Labeling Discrepancy Detection in Surgical Pathology Accession and Grossing Workflows

• Nguyen Minh Kiet

  Faculty of Information Technology, Quy Nhon University, 170 An Duong Vuong Street, Nguyen Van Cu Ward, Quy Nhon City 55100, Binh Dinh Province, Vietnam

  Author

• Tran Quoc Bao

  Department of Computer Science, Thai Binh University, Tan Binh Ward, Thai Binh City 410000, Thai Binh Province, Vietnam

  Author

• Le Duc Thang

  Faculty of Engineering and Technology, Hong Duc University, 565 Quang Trung Street, Dong Ve Ward, Thanh Hoa City 440000, Thanh Hoa Province, Vietnam

  Author

Surgical pathology depends on the correct linkage between patient identity, specimen container, requisition text, tissue site, laterality, procedure description, and gross-room handling. A discrepancy at accession or grossing can delay diagnosis, require recollection, create amended reports, or in rare circumstances place a patient at risk of an incorrect result being associated with the wrong body site or specimen. Automated safety review is possible because accession systems contain structured fields and free-text requisitions, yet the language and workflow patterns differ across services and hospitals. This paper reports an empirical study of service-specific micro adaptation for detecting specimen labeling discrepancies before final pathology sign-out. We assembled 128,740 surgical pathology cases from three hospitals and manually adjudicated 9,860 discrepant or high-risk cases into five discrepancy categories: patient-identifier mismatch, container-requisition mismatch, site or laterality conflict, specimen-count inconsistency, and unsupported free-text amendment. A general transformer classifier was compared with a micro-adapted model that uses small service-specific adapter modules, accession-field consistency checks, and gross-room event features. The proposed model improved macro F1 from 0.692 to 0.801, increased recall for site or laterality conflict from 70.5% to 86.9%, and reduced false safety holds from 11.8 to 7.2 per 1,000 cases. Cross-hospital temporal testing showed smaller degradation than a single global model. Error review indicated that micro adaptation helped most in breast, dermatopathology, gastrointestinal, and gynecologic workflows, where terminology and container conventions differed substantially.

• PDF