报告详情
NarrTBI: A Transformer Encoder for Detecting Sports-Related Head Injury from Emergency Department Narratives in the NEISS Database
编号：53
访问权限：仅限参会人
更新：2026-11-10 16:32:23 浏览：8次
In-person
报告开始：2026年11月20日 14:30（Asia/Kolkata）
报告时间：15min
所在会场：[S4] Artificial Intelligence and Intelligent Computing [S4-1] Artificial Intelligence and Intelligent Computing
暂无文件
摘要
Surveillance of sport-related traumatic brain injury (TBI) depends on diagnosis and body-part codes that clinicians assign at the emergency department, while the free-text narrative written at the same visit is rarely used. We tested how well a Transformer encoder trained from scratch can recognise coded TBI (concussion or internal head injury) from the narrative alone. The data are 75,636 sports and recreation narratives from the 2025 National Electronic Injury Surveillance System, of which 8,040 (10.6%) carry the CDC-style TBI code pair. The diagnosis statement that clinicians append to 99.9% of narratives was cut before training, because an earlier run that removed it only when followed by a colon left the label partly readable. On 11,346 held-out narratives the Transformer reached ROC-AUC 0.966 (95% CI 0.962–0.970) and PR-AUC 0.798 (0.775–0.821). A TF-IDF logistic regression scored slightly higher (0.972 and 0.808); the paired ROC-AUC difference was −0.006 (−0.008 to −0.004). Masking explicit TBI words lowered the Transformer only to 0.963, so the signal lies mainly in how the injury is described. At 90% specificity the Transformer found 89.4% of coded TBI.
关键词
traumatic brain injury, concussion, clinical narratives, NEISS, Transformer encoder, text classification, injury surveillance
报告人
BINTHU MATHAVAN
ASSISTANT PROFESSOR Chennai；CENTRAL UNIVERSITY OF TAMILNADU稿件作者
全部评论
重要日期
-
会议日期
11月20日
2026
至11月22日
2026
-
10月31日 2026
初稿截稿日期
-
11月07日 2026
提前注册日期
-
11月22日 2026
注册截止日期
主办单位
United Societies of Science
承办单位
Pondicherry University
协办单位
IEEE Section
IEEE Madras Section
IEEE Madras Section
联系方式
- ICon INDIA 2026
- ic******@usssociety.org
发表评论