The predictive accuracy of CT radiomics combined with machine learning in predicting the invasiveness of small nodular lung adenocarcinoma

Rong-Sheng Liu; Jia Ye; Yang Yu; Zhi-Yan Yang; Jun-Lv Lin; Xiao-Dong Li; Tian-Shou Qin; Da-Peng Tao; Wei Song; Gang Wang; Jun Peng

doi:10.21037/tlcr-23-82

The predictive accuracy of CT radiomics combined with machine learning in predicting the invasiveness of small nodular lung adenocarcinoma

Transl Lung Cancer Res. 2023 Mar 31;12(3):530-546. doi: 10.21037/tlcr-23-82. Epub 2023 Mar 24.

Authors

Rong-Sheng Liu^#¹, Jia Ye^#¹, Yang Yu², Zhi-Yan Yang¹, Jun-Lv Lin², Xiao-Dong Li¹, Tian-Shou Qin¹, Da-Peng Tao³, Wei Song⁴, Gang Wang⁴, Jun Peng²

Affiliations

¹ Medical School, Kunming University of Science and Technology, Kunming, China.
² Department of Thoracic Surgery, The First People's Hospital of Yunnan Province, The Affiliated Hospital of Kunming University of Science and Technology, Kunming, China.
³ School of Information Science and Engineering, Yunnan University, Kunming, China.
⁴ Department of Radiology, The First People's Hospital of Yunnan Province, The Affiliated Hospital of Kunming University of Science and Technology, Kunming, China.

^# Contributed equally.

Abstract

Background: Conventionally, the judgment of whether small pulmonary nodules are invasive is mainly made by thoracic surgeons according to the chest computed tomography (CT) features of patients. However, there are limits to how much useful information can be obtained from this approach. A large number of feature information was extracted from CT images by CT radiomics. The machine learning algorithm was used to construct models based on radiomic characteristics to predict the invasiveness of lung adenocarcinoma (LUAD) with a good prediction accuracy.

Methods: A total of 416 patients with pathologically confirmed preinvasive lesions and LUAD after video-assisted thoracoscopic surgery (VATS) in the Department of Thoracic Surgery of the First People's Hospital of Yunnan Province from February 2020 to February 2022 were retrospectively analyzed. According to random classification, patients were divided into 2 groups. The RadCloud platform was used to extract radiomics features, and the most relevant radiomics features were selected by continuous dimension reduction method. Then, 6 machine learning algorithms were used to establish and verify the prediction model of small lung nodular adenocarcinoma invasiveness. Receiver operating characteristic (ROC) curve and area under curve (AUC) were used to evaluate the predictive performance.

Results: There were 78 cases of pre-invasive lesions and 226 cases of invasive lesions in the training group, and 34 cases of pre-invasive lesions and 78 cases of invasive lesions in the validation group. In the training group, the AUC values of the 6 models were all more than 0.914, the 95% confidence interval (CI) was 0.857-1.00, the sensitivity was equal or more than 0.87, and the specificity was equal or more than 0.85. In the validation group, the AUC values of the 6 models were all equal or more than 0.732, the 95% CI was 0.651-1.00, the sensitivity was equal or more than 0.7, and the specificity was more than 0.77.

Conclusions: Machine learning algorithms were used to construct models to predict the invasiveness of small nodular LUAD based on radiomics features, which it could provide more evidence for doctors to make diagnoses and more personalized treatment plans for patients.

Keywords: Radiomics; invasiveness; lung adenocarcinoma (LUAD); machine learning; prediction model.