Kwak Se Hyun, Kim Kyeong Yeon, Choi Ji Soo, Kim Min Chul, Seol Chang Hwan, Kim Sung Ryeol, Lee Eun Hye
Division of Pulmonology, Allergy and Critical Care Medicine, Department of Internal Medicine, Yongin Severance Hospital, Yonsei University College of Medicine, Yongin-si, Republic of Korea.
Center for Digital Health, Yongin Severance Hospital, Yonsei University College of Medicine, Yongin-si, Gyeonggi-do, Republic of Korea.
Front Med (Lausanne). 2024 Aug 7;11:1449537. doi: 10.3389/fmed.2024.1449537. eCollection 2024.
The use of artificial intelligence (AI) for chest X-ray (CXR) analysis is becoming increasingly prevalent in medical environments. This study aimed to determine whether AI in CXR can unexpectedly detect lung nodule detection and influence patient diagnosis and management in non-respiratory outpatient clinics.
In this retrospective study, patients over 18 years of age, who underwent CXR at Yongin Severance Hospital outpatient clinics between March 2021 and January 2023 and were identified to have lung nodules through AI software, were included. Commercially available AI-based lesion detection software (Lunit INSIGHT CXR) was used to detect lung nodules.
Out Of 56,802 radiographic procedures, 40,191 were from non-respiratory departments, with AI detecting lung nodules in 1,754 cases (4.4%). Excluding 139 patients with known lung lesions, 1,615 patients were included in the final analysis. Out of these, 30.7% (495/1,615) underwent respiratory consultation and 31.7% underwent chest CT scans (512/1,615). As a result of the CT scans, 71.5% (366 cases) were found to have true nodules. Among these, the final diagnoses included 36 lung cancers (7.0%, 36/512), 141 lung nodules requiring follow-up (27.5%, 141/512), 114 active pulmonary infections (22.3%, 114/512), and 75 old inflammatory sequelae (14.6%, 75/512). The mean AI nodule score for lung cancer was significantly higher than that for other nodules (56.72 vs. 33.44, < 0.001). Additionally, active pulmonary infection had a higher consolidation score, and old inflammatory sequelae had the highest fibrosis score, demonstrating differences in the AI analysis among the final diagnosis groups.
This study indicates that AI-detected incidental nodule abnormalities on CXR in non-respiratory outpatient clinics result in a substantial number of clinically significant diagnoses, emphasizing AI's role in detecting lung nodules and need for further evaluation and specialist consultation for proper diagnosis and management.
在医疗环境中，利用人工智能（AI）进行胸部X光（CXR）分析正变得越来越普遍。本研究旨在确定CXR中的AI是否能意外检测到肺结节，并影响非呼吸科门诊患者的诊断和管理。
在这项回顾性研究中，纳入了2021年3月至2023年1月在龙仁圣母医院门诊接受CXR检查且通过AI软件被识别出有肺结节的18岁以上患者。使用市售的基于AI的病变检测软件（Lunit INSIGHT CXR）来检测肺结节。
在56,802例放射检查中，40,191例来自非呼吸科，AI在1,754例（4.4%）中检测到肺结节。排除139例已知肺部病变的患者后，1,615例患者纳入最终分析。其中，30.7%（495/1,615）接受了呼吸科会诊，31.7%接受了胸部CT扫描（512/1,615）。CT扫描结果显示，71.5%（366例）有真正的结节。其中，最终诊断包括36例肺癌（7.0%，36/512）、141例需要随访的肺结节（27.5%，141/512）、114例活动性肺部感染（22.3%，114/512）和75例陈旧性炎症后遗症（14.6%，75/512）。肺癌的平均AI结节评分显著高于其他结节（56.72对33.44，<0.001）。此外，活动性肺部感染的实变评分更高，陈旧性炎症后遗症的纤维化评分最高，表明最终诊断组之间的AI分析存在差异。
本研究表明，非呼吸科门诊中AI检测到的CXR偶然结节异常导致了大量具有临床意义的诊断，强调了AI在检测肺结节中的作用以及对正确诊断和管理进行进一步评估和专科会诊的必要性。
