Development and evaluation of a venous computed tomography radiomics model to predict lymph node metastasis from non-small cell lung cancer. Issue 18 (May 2020)

• Record Type: Journal Article
• Authors: Cong, Mengdi; Yao, Haoyue; Liu, Hui; Huang, Liqiang; Shi, Gaofeng
• Other Names: Lalwani. Neeraj section editor.
• Is Part Of: Medicine. Volume 99:Issue 18(2020)
• Publication Date: 2020-05
• Subjects: forecasting -- lymph nodes -- non-small cell lung cancer -- radiomics -- tomography -- X-ray computed; Medicine -- Periodicals; Medicine -- Periodicals; Médecine -- Périodiques; Geneeskunde; Medicine; Periodicals; Periodicals; 610.5
• DOI: 10.1097/MD.0000000000020074
• Languages: English
• ISSNs: 0025-7974

Abstract

Abstract : Abstract: The objective of this study was to develop a venous computed tomography (CT)-based radiomics model to predict the lymph node metastasis (LNM) in patients with non-small cell lung cancer (NSCLC). A total of 411 consecutive patients with NSCLC underwent tumor resection and lymph node (LN) dissection from January 2018 to September 2018 in our hospital. A radiologist with 20 years of diagnostic experience retrospectively reviewed all CT scans and classified all visible LNs into LNM and non-LNM groups without the knowledge of pathological diagnosis. A logistic regression model (radiomics model) in classification of pathology-confirmed NSCLC patients with and without LNM was developed on radiomics features for NSCLC patients. A morphology model was also developed on qualitative morphology features in venous CT scans. A training group included 288 patients (99 with and 189 without LNM) and a validation group included 123 patients (42 and 81, respectively). The receiver operating characteristic curve was performed to discriminate LNM (+) from LNM (−) for CT-reported status, the morphology model and the radiomics model. The area under the curve value in LNM classification on the training group was significantly greater at 0.79 (95% confidence interval [CI]: 0.77–0.81) by use of the radiomics model (build by best 10 features in predicting LNM) compared with 0.51 by CT-reported LN status ( P < .001) or 0.66 (95% CI: 0.64–0.68) by morphology model (build by tumor size and spiculation) ( P < .001). Similarly, the area under the curve value on the validation group was 0.73 (95% CI: 0.70–0.76) by the radiomics model, compared with 0.52 or 0.63 (95% CI: 0.60–0.66) by the other 2 (both P < .001). A radiomics model shows excellent performance for predicting LNM in NSCLC patients. This predictive radiomics model may benefit patients to get better treatments such as an appropriate surgery.