An explainable machine learning model combining intratumoral and peritumoral computed tomography radiomics with maximum standard uptake value for predicting International Association for the Study of Lung Cancer grading and prognosis in stage I pure-solid lung adenocarcinoma.
Authors
Affiliations (5)
Affiliations (5)
- Department of Radiology, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China.
- Department of Radiology, Jinhua Center Hospital, Zhejiang University School of Medicine, Jinhua, China.
- Department of Radiology, The Quzhou Affiliated Hospital of Wenzhou Medical University, Quzhou People's Hospital, Quzhou, China.
- Department of Nuclear Medicine, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China.
- Department of Research and Development, Shanghai United Imaging Intelligence Co., Ltd., Shanghai, China.
Abstract
The novel International Association for the Study of Lung Cancer (IASLC) grading system provides important prognostic information for lung adenocarcinoma; however, preoperative noninvasive prediction of this grading remains challenging. This study aimed to evaluate the predictive value of a combined intratumoral and peritumoral radiomics model based on computed tomography (CT) and maximum standard uptake value (SUV<sub>max</sub>) for preoperative prediction of the novel IASLC grading and prognosis in stage I pure-solid invasive lung adenocarcinoma (LADC). Patients with stage I pure-solid LADC from The First Affiliated Hospital of Wenzhou Medical University were retrospectively included and randomly assigned to a training set (n=207) and a testing set (n=89) in a 7:3 ratio. There were 851 radiomics features extracted from the gross tumor volume (GTV) and the GTV incorporating peritumoral 3, 6, and 9 mm regions (GPTV3, GPTV6, GPTV9). A combined model based on an individual radiomics signature, SUV<sub>max</sub>, and clinical independent predictors was constructed. Recurrence-free survival (RFS) was analyzed using Kaplan-Meier survival method, while independent prognostic factors were identified through Cox proportional hazards models. The GPTV3 model demonstrated a higher area under the curve (AUC) compared to other individual radiomics models, with AUCs of 0.804 and 0.715 in the training and testing sets. Tumor size was identified as an independent predictor of high-grade IASLC grading in the clinical model. The combined model showed AUCs of 0.837 and 0.767 for the training and testing sets, respectively, which were higher than those of the individual GPTV3 and SUV<sub>max</sub> models. The high-risk group was associated with poorer RFS (P<0.05). Pathological and predictive IASLC grade, smoking history, and tumor size were independent prognostic factors. The interpretable machine learning model can predict the novel IASLC grading and prognosis in stage I pure-solid LADC.