Objective To explore the predictive value of artificial intelligence (AI)-based lung nodule CT quantitative analysis for the invasion degree of lung adenocarcinoma spectrum lesions. Methods According to the invasion degree of lung adenocarcinoma spectrum lesions, patients with surgically and pathologically confirmed lung adenocarcinoma spectrum lesions from January 2023 to June 2023 in Jinling Hospital affiliated to Nanjing University Medical School were retrospectively collected and divided into a non-invasive group and an invasive group, including atypical adenomatous hyperplasia, adenocarcinoma in situ, and micro-invasive adenocarcinoma patients in the non-invasive group, and invasive adenocarcinoma patients in the invasive group. All enrolled patients underwent chest CT imaging before surgery, and then the lung nodules were quantitatively analyzed using an AI-based computer-aided diagnosis system to compare the related quantitative parameters of lung nodules that have been surgically removed and pathologically confirmed as lung adenocarcinoma spectrum lesions between the two groups, and to analyze the relationship between various CT quantitative features and the invasion degree of lung adenocarcinoma spectrum lesions. Results A total of 149 patients (149 lesions) were included, including 42 males and 107 females, aged 29-81 (56.35±10.75) years. There were 72 patients in the non-invasive group and 77 patients in the invasive group. The differences in long diameter, short diameter, volume, surface area, mass, maximum surface area, 3D long diameter, maximum CT value, minimum CT value, average CT value, entropy, kurtosis, skewness, and malignancy probability of lung nodules between the two groups were statistically significant (P<0.05). Multivariate binary logistic regression analysis showed that long diameter [OR=1.687, 95%CI (1.364, 2.085), P<0.001], average CT value [OR=1.006, 95%CI (1.002, 1.009), P=0.002], and malignancy probability [OR=1.034, 95%CI (1.005, 1.063), P=0.020] were independent risk factors for aggravating the invasion degree of lung adenocarcinoma. The predictive model combining the above parameters demonstrated optimal performance, with an area under the receiver operating characteristic curve of 0.951, sensitivity of 0.818, and specificity of 0.972. Using a Nomogram to quantify the three independent risk factors, the cross-validation was performed to evaluate the stability of the model, and the average C-index of cross-validation was 0.950, with each fold C-index >0.75, indicating that the prediction performance of the model was stable, and the calibration curve and decision curve indicated good predictive performance. Conclusion The visualization prediction model constructed by AI-based quantitative analysis of lung nodules in CT demonstrates significant discriminative effectiveness in the assessment of invasiveness in lung adenocarcinoma spectrum lesions. This visualization prediction model can provide a quantitative decision-making basis for the preoperative identification of the degree of invasiveness in lung adenocarcinoma spectrum lesions.

Copyright © the editorial department of Chinese Journal of Clinical Thoracic and Cardiovascular Surgery of West China Medical Publisher. All rights reserved