AI肺小结节检测系统用于低剂量CT肺筛查中的应用研究

Application of AI Pulmonar y Nodule Detection System in Low-dose CT Lung Screening

目的 探讨人工智能(AI)肺小结节检测系统用于低剂量CT肺筛查中的应用价值。方法 回顾性分析广东医科大学附属第二医院在2020年12月至2022年6月期间开展低剂量CT筛查的肺小结节患者150例的临床资料,将获取的CT图像分别用人工检测法、AI肺小结节检测系统进行阅片,记录结节检出率。同时以超声支气管镜引导下经支气管肺活检术(EBUSTBLB)检查为金标准,比较不同方法对恶性结节的检出价值。结果 150例肺筛查患者以联合应用为标准,共筛查出154个肺小结节,AI组检出数186个高于人工组的113个及联合应用组的154个,漏检率0低于人工组的36.28%,但误检率17.20%高于人工组的1.7 7%,差异有统计学意义(P<0.05)。AI组对直径<5mm、5~10mm的结节检出率高于人工组,肺周围区的肺小结节检出率高于对照组(P<0.05),两组对直径>10mm的结节检出率比较,差异无统计学意义(P>0.05)。AI检测组的阅片时间、平均检出时间均短于人工组(P<0.05)。150例患者以EBUS-TBLB检查结果为金标准,其中恶性结节11例(7.33%),联合应用对恶性肺小结节的诊断灵敏度、准确性、阳性预测值均高于人工检测、AI肺小结节检测,差异有统计学意义(P<0.05)。结论 AI肺小结节检测系统用于低剂量CT肺筛查,肺小结节检出率较高,减少阅片时间,提高阅片效率,但人工阅片辅助AI阅片,可降低疾病漏诊率,有效鉴别病灶良恶性。

Objective To explore the application value of artificial intelligence(AI)pulmonary nodule detection system for low-dose CT lung screening.Methods The clinical data of 150 patients with small pulmonary nodules who underwent low-dose CT screening from December 2020 to June 2022 in the Second Affiliated Hospital of Guangdong Medical University were retrospectively analyzed.The AI pulmonary nodule detection system reads the images and records the nodule detection rate.At the same time,ultrasound bronchoscopy-guided transbronchial lung biopsy(EBUS-TBLB)was used as the gold standard to compare the detection value of different methods for malignant nodules.Results A total of 154 small pulmonary nodules were screened in 150 lung screening patients with the combined application as the standard.The number of 186 detected in the AI group was higher than 113 in the manual group and 154 in the combined application group,and the missed detection rate was 0.It was lower than 36.28%of the manual group,but the false detection rate was 17.20%higher than that of the manual group,1.77%,and the difference was statistically significant(P<0.05).The detection rate of nodules with diameters<5 mm and 5-10 mm in the AI group was higher than that in the artificial group,and the detection rate of pulmonary nodules in the peripheral area of the lung was higher than that in the control group(P<0.05),there was no significant difference in the detection rate of nodules with a diameter of more than 10 mm between the two groups(P>0.05).The reading time and average detection time of the AI detection group were shorter than those of the manual group(P<0.05).The results of EBUS-TBLB examination were used as the gold standard in 150 patients,of which 11 were malignant nodules(7.33%).In the detection of small nodules,the difference was statistically significant(P<0.05).Conclusion The AI pulmonary nodule detection system is used for low-dose CT lung screening,and the detection rate of small pulmonary nodules is high,reducing the reading time and improving the efficiency of reading.It can effectively distinguish benign and malignant lesions.