A comparative analysis of peripheral lung shadow diagnosed by artificial intelligence-assisted chest CT versus postoperative pathology in 810 patients
- VernacularTitle:810例周边型肺阴影的人工智能辅助胸部CT诊断与术后病理诊断对比分析
- Author:
Lin DU
1
;
Hong ZHANG
1
;
Xiangfeng LUO
2
;
Jun LV
1
;
Daqiang SUN
1
Author Information
1. Department of Thoracic Surgery, Tianjin Chest Hospital Affiliated to Tianjin University, Tianjin, 300222, P. R. China
2. LinkDoc Science and Technology Co. Ltd., Beijing, 100080, P. R. China
- Publication Type:Journal Article
- Keywords:
Lung cancer;
artificial intelligence;
auxiliary diagnosis;
peripheral lung shadow
- From:
Chinese Journal of Clinical Thoracic and Cardiovascular Surgery
2022;29(07):854-858
- CountryChina
- Language:Chinese
-
Abstract:
Objective To investigate the clinical value of artificial intelligence (AI)-assisted chest computed tomography (CT) in the diagnosis of peripheral lung shadow. Methods The CT image data of 810 patients with peripheral pulmonary shadow treated by thoracic surgery in Tianjin Chest Hospital Affiliated to Tianjin University from January 2018 to July 2019 were retrospectively analyzed using AI-assisted chest CT imaging diagnosis system. There were 339 males and 471 females with a median age of 63 years. The malignant probability of preoperative AI-assisted diagnosis of peripheral pulmonary shadow was compared with the results of postoperative pathology. Results The pathological diagnosis of 810 patients with peripheral pulmonary shadow was lung cancer in 627 (77.4%) patients, precancerous lesion in 30 (3.7%) patients and benign lesion in 153 (18.9%) patients. The median probability of malignant AI diagnosis before operation was 86.0% (lung cancer), 90.0% (precancerous lesion) and 37.0% (benign lesion), respectively. According to the analysis of receiver operating characteristic (ROC) curve of AI malignant probability distribution in this group of patients, the area under the ROC curve was 0.882. The critical value of malignant probability for diagnosis of lung cancer was 75.0%with a sensitivity of 0.856 and specificity of 0.814. A total of 571 patients were diagnosed with AI malignancy probability≥ 75.0%, among whom 537 patients were pathologically diagnosed as lung cancer with a positive predictive value of 94.0%(537/571). Conclusion The AI-assisted chest CT diagnosis system has a high accuracy in the diagnosis of peripheral lung cancer with malignant probability≥75.0% as the diagnostic threshold.