目的

基于人工智能（artificial intelligence，AI）的CT 定量分析在肺结节性质判断及浸润程度中的价值。

方法

回顾性分析2021 年9 月至2023 年5 月在我院就诊的182 例经手术病理确诊为肺癌的患者，其中浸润性腺癌（minimally invasive adenocarcinoma，MIC）63 例、微浸润性腺癌（minimally invasive adenocarcinoma，MIA）104 例、原位癌（adenocarcinoma in situ，AIS）15 例，行AI 的CT 定量分析。对比病理和AI 辅助CT 检查对于肺结节性质和浸润程度的诊断，同时对比实性肺结节和磨玻璃肺结节、不同浸润程度的肺结节AI 分析参数。

结果

通过AI 软件对182 例患者的CT 图像进行定量分析，其中95 例为实性结节，87 例为磨玻璃结节；15 例为原位肺结节，106 例为微浸润肺结节，63 例为浸润肺结节。AI 辅助CT 检查诊断肺结节性质判定和浸润程度的灵敏度、特异度、准确性、阳性和阴性预测值与组织病理学诊断比较，无统计学意义（P＞0.05）；实性结节与磨玻璃结节CT 参数对比显示：实性结节质量为（448.34±60.05）g，3D 长径为（16.88±0.80）mm、长短径平均值为（12.99±0.63）mm，CT 平均值为（-453.29±15.17）Hu，CT 方差为（107 460.12±17 115.26）Hu，CT 最大值为（401.03±32.37）Hu，熵值为（9.09±0.88），高于磨玻璃结节（P＜0.05）；不同浸润程度肺结节CT 参数显示：原位润肺结节质量为（238.04±163.23）g，体积为（716.66±479.08）cm³，高于浸润癌，但低于微浸润癌，3D 长径和CT 平均值均低于微浸润癌和浸润癌（P＜0.05）；与原位癌比较，微浸润癌长短经平均值高，浸润癌长短经平均值低（P＜0.05）。

结论

AI 辅助CT 检查可准确诊断肺结节性质与浸润程度，降低误诊与漏诊。

Objective

To evaluate the value of AI-based CT quantitative analysis on pulmonary nodules.

Methods

A total of 182 patients with lung cancer confirmed by surgery and pathology in Xinqiao Hospital from September 2021 to May 2022 were retrospectively analyzed，including 63 cases of invasive adenocarcinoma （MIC），104 cases of minimally invasive adenocarcinoma （MIA） and 15 cases of carcinoma in situ （AIS）. All patients underwent CT quantitative analysis of AI. The diagnostic efficacy of pathology and AI-assisted CT examination for the nature and infiltration degree of pulmonary nodules was compared. At the same time，the AI analysis parameters of solid pulmonary nodules and ground glass pulmonary nodules and different infiltration degree of pulmonary nodules were compared.

Results

Surgical pathology showed that 95 cases were solid nodules，87 were ground glass nodules；15 cases were in situ pulmonary nodules，106 were microinvasive pulmonary nodules and 63 were invasive pulmonary nodules. The sensitivity，specificity，accuracy，positive predictive value and negative predictive value of the gold standard，No statistical difference （P＞0.05）；Comparison of CT parameters between solid nodules and ground glass nodules showed that：The mass of solid nodules was （448.34±60.05） g，the 3D long diameter was （16.88±0.80） mm，the average length and short diameter was （12.99±0.63）mm，the average CT value was （-453.29±15.17）Hu，and the CT variance was（107 460.12±17 115.26） Hu. The maximum value of CT was （401.03±32.37） Hu，and the entropy value was（9.09±0.88），which were higher than those of ground glass nodules，and there were differences between the groups （P＜0.05）. The CT parameters of lung nodules with different degrees of invasion showed that the weight and volume of lung nodules in situ were （238.04±163.23） g and （716.66±479.08） cm³，which were higher than those of invasive carcinoma，but lower than those of minimally invasive carcinoma，and the 3D long diameter and CT average were lower than those of minimally invasive carcinoma and invasive carcinoma （P＜0.05）. Compared with carcinoma in situ，microinvasive carcinoma had a higher mean length of meridian and invasive carcinoma had a lower mean length of meridian （P ＜0.05）.

Conclusion

The AI-assisted CT examination can accurately diagnose the nature and infiltration degree of pulmonary nodules，and the quantitative analysis of relevant parameters can provide a basis for disease evaluation.