深度学习算法在间质性肺病胸片网状阴影检测中的应用价值
本回顾性研究纳入197例经手术证实间质性肺病患者及197例匹配对照,评估商用深度学习算法(DLA)检测胸片网状阴影的效能。DLA的敏感度、特异度、准确率分别为98.0%、99.0%、98.5%,均显著优于独立阅片医生(77.3%、92.3%、84.8%),且医生使用DLA后各指标及观察者间一致性均显著提高。DLA在轻度病变中仍保持较高敏感度,并在第二机构102张胸片上达到100.0%准确率。
要点
- DLA检测网状阴影的敏感度、特异度、准确率均显著高于独立阅片医生。
- 医生借助DLA可将敏感度从77.3%提升至93.8%,观察者间一致性从κ=0.517提升至κ=0.870。
- DLA在轻度病变中敏感度仍达86.8%,有助于早期间质性肺病的检出。
局限性
- 回顾性研究设计。
- 样本来自单中心及第二机构较小样本验证,需多中心前瞻性研究进一步确认。
- DLA为商用算法,其泛化性受设备及参数影响。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
BACKGROUND. Deep learning has been heavily explored for pulmonary nodule detection on chest radiographs. Detection of reticular opacity in interstitial lung disease (ILD) is challenging and may also benefit from a deep learning algorithm (DLA). OBJECTIVE. The purpose of this study was to evaluate the utility of a DLA for detection of reticular opacity on chest radiographs of patients with surgically confirmed ILD. METHODS. This retrospective study included 197 patients (130 men, 67 women; mean age, 62.6 ± 7.6 [SD] years) with surgically proven ILD between January 2017 and December 2018 who underwent preoperative chest radiography and chest CT within a 30-day interval. A total of 197 age- and sex-matched control patients with normal chest radiographs were randomly selected. A commercially available DLA was used to detect lower lobe or subpleural abnormalities; those matching the reticular opacity location on CT were deemed true-positive. Six readers (three thoracic radiologists, three residents) independently reviewed radiographs with and without the DLA for the presence of reticular opacity. Interobserver agreement was assessed. Diagnostic performance was compared among interpretations. Subanalysis was performed according to CT-based classification of the severity of reticular opacity. Performance of the DLA was also assessed on 102 chest radiographs from a second institution (51 patients with ILD, 51 matched patients in the control group). RESULTS. Interobserver agreement was moderate (κ = 0.517) for readers alone and almost perfect (κ = 0.870) for readers using the DLA. Sensitivity, specificity, and accuracy of the DLA for reticular opacity were 98.0%, 99.0%, and 98.5%; of pooled readers alone were 77.3%, 92.3%, and 84.8%; and of readers using the DLA were 93.8%, 97.3%, and 95.6%. All metrics were significantly better (all p ≤ .002) for the DLA and for readers using the DLA than for readers alone. Sensitivity for readers without and with the DLA were 66.7% and 86.8% for mild disease, 84.2% and 98.8% for moderate disease, and 87.3% and 100.0% for severe disease. The DLA had 100.0% accuracy at the second institution. CONCLUSION. The DLA outperformed readers in detection of reticular opacity, and use of the DLA improved reader performance and interobserver agreement. The benefit of the DLA was more notable in sensitivity than in specificity and was maintained in mild disease. CLINICAL IMPACT. Use of the DLA may facilitate detection of reticular opacity on chest radiographs in the early stages of ILD.
引用
- Kim W, Lee SM, Kim JI, et al. Utility of a Deep Learning Algorithm for Detection of Reticular Opacity on Chest Radiography in Patients With Interstitial Lung Disease. AJR Am J Roentgenol. 2021. PMID: 34668769. doi:10.2214/AJR.21.26682
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。