基于增强CT影像的人工智能模型在肺腺癌浸润性评估中的诊断价值

Diagnostic value of an artificial intelligence model based on contrast-enhanced CT imaging in the assessment of lung adenocarcinoma invasiveness

摘要:

目的 探讨基于增强CT影像的人工智能（AI）模型在肺腺癌浸润性评估中的诊断价值。方法 选取2022年1月—2025年1月在宿州市立医院就诊、胸部CT发现肺结节并计划接受手术切除的患者为研究对象。所有患者术前接受增强CT扫描，并将扫描图像实时导入AI辅助诊断系统进行分析，由系统自动生成关于结节是否为浸润性肺腺癌的预测概率及量化参数。以术后病理诊断为金标准，评估AI辅助CT增强扫描对肺腺癌浸润性的术前预测效能。结果 141例患者经手术证实为肺腺癌的患者中，浸润性腺癌组为68.09%（96/141）。浸润性腺癌组与非浸润性腺癌组的年龄、体质量指数、性别构成、吸烟史、饮酒史、居住地构成、文化程度构成比较，差异均无统计学意义（P >0.05）。浸润性腺癌组的体积、质量、最大CT值、最小CT值、CT值标准差、最大层面积、3D长径、长短径平均值和量化参数熵均大于非浸润性腺癌组（P <0.05）。141例患者中浸润性为96例，增强CT扫描诊断阳性79例，AI辅助CT增强诊断阳性92例。增强CT扫描诊断肺腺癌浸润性的敏感性为82.29%（95% CI：0.732，0.893）（79/96），特异性为75.56%（95% CI：0.605，0.871）（34/45），准确性为80.14%（95% CI：0.726，0.861）（113/141）；AI辅助CT增强检测肺腺癌浸润性的敏感性为95.83%（95% CI：0.897，0.989）（92/96），特异性为88.89%（95% CI：0.759，0.963）（40/45），准确性为93.62%（95% CI：0.880，0.970）（132/141）。结论 AI联合CT增强通过精准量化肺结节相关参数，可提升肺腺癌肺浸润性诊断的准确性、敏感性和特异性。

Abstract:

Objective To investigate the diagnostic value of an artificial intelligence (AI) model based on contrast-enhanced CT imaging in the assessment of lung adenocarcinoma invasiveness.Methods Patients who visited Suzhou Municipal Hospital between January 2022 and January 2025, presented with pulmonary nodules on chest CT scans, and were scheduled for surgical resection were enrolled as the study subjects. All patients underwent contrast-enhanced CT scans preoperatively, and the scanned images were immediately imported into an AI-assisted diagnostic system for analysis. The system automatically generated the predicted probability of whether the nodule was an invasive lung adenocarcinoma, along with quantitative parameters. Using postoperative pathological diagnosis as the gold standard, the preoperative predictive efficacy of AI-assisted contrast-enhanced CT scans for assessing the invasiveness of lung adenocarcinoma was evaluated.Results Among 141 patients confirmed with lung adenocarcinoma by surgery, the invasive group accounted for 68.09% (96/141). The comparisons of age, BMI, gender composition, smoking history rate, drinking history rate, residential composition, educational level composition, and kurtosis between the invasive and non-invasive groups showed no statistically significant differences (P > 0.05). In contrast, comparisons of volume, mass, maximum CT value, minimum CT value, standard deviation of CT value, maximum cross-sectional area, 3D longest diameter, average long-short axis diameter, and the quantitative parameter entropy in the invasive group showed statistically significant differences (P < 0.05), with all values higher in the invasive group than in the non-invasive group. Among the 141 patients, 96 were classified as invasive. Conventional contrast-enhanced CT detected positivity in 79 cases, while AI-assisted contrast-enhanced CT detected positivity in 92 cases. The sensitivity of contrast-enhanced CT scan for diagnosing the invasiveness of lung adenocarcinoma was 82.29% (95% CI: 0.732, 0.893) (79/96), the specificity was 75.56% (95% CI: 0.605, 0.871) (34/45), and the accuracy was 80.14% (95% CI: 0.726, 0.861) (113/141). The sensitivity of AI-assisted contrast-enhanced CT for detecting the invasiveness of lung adenocarcinoma was 95.83% (95% CI: 0.897, 0.989) (92/96), the specificity was 88.89% (95% CI: 0.759, 0.963) (40/45), and the accuracy was 93.62% (95% CI: 0.880, 0.970) (132/141).Conclusion The combination of artificial intelligence and contrast-enhanced CT enables the accurate quantification of parameters related to pulmonary nodules, thereby improving the accuracy, sensitivity, and specificity in the diagnosis of lung adenocarcinoma invasiveness.