PO.BCS02.02 · 生物信息与计算
利用人工智能最大化高危偶发肺结节的转诊
Maximizing high-risk incidental pulmonary nodule referrals using artificial intelligence
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
背景：有效的肺癌筛查和肺结节（PN）管理项目需要及时的检测和稳健的诊疗协调。我们报告一款人工智能（AI）工具在检测到高危PN征象后用于获取肺科转诊的初步结果。
方法：JPS Health Network采用国家公认的PN指南创建了肺结节风险评分（PNRS）分类系统。AI驱动的自然语言处理（NLP）使导航员能够识别符合PNRS 4B标准（表1）——即最高风险PN类别——的患者。对NLP实施前后各11个月、共22个月期间的PNRS 4B检出率和转诊率进行了比较。统计分析对分类变量采用Fisher精确检验，对连续变量采用Mann-Whitney U检验或t检验。
结果：2023年6月1日至2024年4月25日（NLP实施前），共进行了26,393例胸部CT；NLP实施后（2024年4月26日至2025年3月20日）获取了23,396例。NLP实施前和实施后分别纳入76例和106例PNRS 4B患者。NLP实施后的工作流程使90天内转诊率从55例（72%）显著提高至98例（93%）（p<0.001）。预约完成率从68%提高至78%（p=0.17）。NLP实施前后比较显示活检完成率无显著差异（65%对60%），恶性肿瘤检出分别为28例（37%）对38例（37%）。若NLP实施前的转诊率能以相同的37%恶性率达到93%，估计将会多诊断出6例患者。
结论：该AI工具显著提高了转诊率并加速了恶性肿瘤的检出，有可能减少高危结节的漏诊或延迟诊断。
PNRS 4B标准及推荐的临床管理路径 PNRS：评分描述：等效的影像学征象：相关推荐：PNRS 4B 高度可疑实性结节：基线时单发或多发≥15 mm（≥1,767 mm³）结节，或新发，或；生长≥8（≥268 mm³）的结节。部分实性结节：基线时实性成分≥8 mm（268 mm³），或新发，或；随访时生长的实性成分≥6 mm（113 mm³）。磨玻璃结节（GGN）：生长≥8 mm（268 mm³）。裂旁或胸膜旁结节：生长≥10 mm（524 mm³）。推荐肺科转诊，且随访诊断性胸部CT，和/或；PET/CT，和/或；组织取样/活检，和/或；GGN考虑切除
查看英文原文 English abstract
Background: Effective lung cancer screening and pulmonary nodule (PN) management programs require timely detection and robust care coordination. We report the initial outcomes of an artificial intelligence (AI) tool for obtaining referrals to pulmonary following the detection of high-risk PN findings.
Methods: JPS Health Network created the Pulmonary Nodule Risk Score (PNRS) classification system using nationally accepted PN guidelines. AI-driven Natural Language Processing (NLP) enabled navigators to identify patients meeting the PNRS 4B criteria (Table 1), the highest risk PN category. PNRS 4B detection and referral rates were compared over 22 months, spanning 11 months pre- and post-NLP implementation. Statistical analysis used Fisher's Exact for categorical variables and Mann-Whitney U or t tests for continuous variables.
Results: From June 1, 2023-April 25, 2024 (pre-NLP), 26,393 chest CTs were performed; 23,396 were obtained post-NLP (April 26, 2024-March 20, 2025). The pre- and post-NLP periods included 76 and 106 PNRS 4B patients, respectively. Post-NLP workflow significantly improved referral within 90 days rates from 55 (72%) to 98 (93%) (p<0.001). Appointment completion improved from 68% to 78% (p=0.17). Pre- and post-NLP comparisons demonstrated no significant differences in biopsy completion (65% vs 60%), Malignancy was noted in 28 (37%) vs 38 (37%). If pre-NLP referrals had reached 93% with the same 37% malignancy rate, an estimated 6 additional patients would have been diagnosed.
Conclusion: The AI tool significantly increased referral rates and expedited malignancy detection, potentially reducing missed or delayed diagnoses in high-risk nodules.
PNRS 4B Criteria and Recommended Clinical Management Pathways PNRS: Score Description: E quivalent Radiology Findings: Associated Recommendation(s): PNRS 4B Very Suspicious Solid Nodule : Single or Multiple ≥ 15 mm (≥ 1,767 mm 3 ) nodule(s) at baseline or new or; Growing ≥ 8 (≥ 268 mm 3 ) nodule. Part Solid Nodule: Solid component ≥ 8 mm (268 mm 3 ) at baseline or new or; Growing solid component ≥ 6 mm (113 mm 3 ) upon follow-up. Ground Glass Nodule (GGN) : Growing ≥ 8 mm (268 mm 3 ). Perifissural or Juxtapleural Nodule : Growing ≥ 10 mm (524 mm 3 ). Pulmonary Referral Recommended AND Follow-Up Diagnostic Chest CT AND/OR; PET/CT AND/OR; Tissue sampling / biopsy AND/OR; GGN Consider Resection
利益披露 Disclosure
I. M. Marrufo, None..
R. Johnson, None..
S. Newman, None..
P. Patel, None..
K. Narra, None.