人工智能辅助肺结节筛查及定性诊断的应用研究

李娟娟; 刘敏; 杨斌; 杜伟; 杨鸿开; 廖艳全; 李峻杭; 王君

doi:10.7619/jcmp.20214698

人工智能辅助肺结节筛查及定性诊断的应用研究

Application of artificial intelligence-assisted pulmonary nodule screening and qualitative diagnosis

摘要

摘要:
目的探讨人工智能(AI)辅助诊断软件在胸部低剂量CT肺结节筛查及定性诊断的临床研究与应用价值。
方法回顾性分析病理确诊的103例肺结节患者的临床资料，将103例肺结节患者术前的胸部低剂量CT图像导入杏脉锐影肺结节AI分析软件中，采用AI及放射医师阅片的方法筛查肺结节并进行良恶性诊断，将AI辅助诊断软件与放射医师对肺结节的筛查情况进行比较，并以病理诊断为金标准，分析AI辅助诊断软件与放射医师诊断的准确性。
结果 103例患者胸部低剂量CT共筛查出258个结节, AI辅助软件与放射医师检出肺结节的敏感度分别为96.12%、89.53%, 阳性预测值分别为95.00%、100.00%; AI辅助诊断软件筛查检出肺结节的假阳性率为5.00%, 放射医师未检查出假阳性肺结节。AI辅助诊断软件与放射医师对肺结节的筛查能力比较，差异有统计学意义(P < 0.05)。103例肺结节患者病理检查共诊断出108个肺结节, AI辅助诊断软件及放射医师诊断肺结节的敏感度分别为95.35%、91.86%，特异度分别为72.73%、81.82%。
结论 AI辅助诊断软件在肺结节筛查检出及恶性结节诊断方面具有较高的准确性，但肺结节良恶性鉴别准确率低于放射医师。因此, AI辅助诊断软件可作为一种辅助手段与放射医师诊断相结合来提高肺结节的总体诊疗效能。

Abstract:
Objective To explore the clinical research and application value of artificial intelligence (AI) aided diagnosis software in lung nodule screening and qualitative diagnosis of CT screening of chest low-dose CT.
Methods The clinical data of 103 patients with pulmonary nodules diagnosed by pathology were analyzed retrospectively. The preoperative chest low-dose CT images of 103 patients with pulmonary nodules were imported into the AI analysis software of apricot pulse sharp shadow pulmonary nodules. The methods of AI and radiologists′film reading were used to screen pulmonary nodules and make benign and malignant diagnosis. The AI aided diagnosis software was compared with the screening of pulmonary nodules by radiologists, and the pathological diagnosis was taken as the gold standard, the accuracy of AI aided diagnosis software and radiologist diagnosis was analyzed.
Results A total of 258 nodules were detected by chest low-dose CT in 103 patients. The sensitivity of pulmonary nodules detected by AI assistant software and radiologist were 96.12% and 89.53%, respectively, the positive predictive values were 95.00% and 100.00%, respectively, the false positive rate of pulmonary nodules detected by AI assisted diagnostic software was 5.00%, and radiologists did not detect false positive pulmonary nodules. There was significant difference between AI aided diagnosis software and radiologists in screening ability of pulmonary nodules (P < 0.05). A total of 108 nodules were diagnosed by pathological examination in 103 patients with pulmonary nodules, the sensitivity of AI aided diagnosis software and radiologists in diagnosing pulmonary nodules were 95.35% and 91.86%, respectively, and the specificities were 72.73% and 81.82%, respectively.
Conclusion AI aided diagnosis software has high accuracy in the screening and detection of pulmonary nodules and the diagnosis of malignant nodules, but the accuracy of differentiating benign from malignant pulmonary nodules is lower than that of radiologists. Therefore, AI aided diagnosis software as an auxiliary approach can be combined with diagnosis of radiologists to improve the overall diagnosis and treatment efficiency of pulmonary nodules.

HTML全文

参考文献(17)

施引文献

资源附件(0)