A Spectral CT-Based Deep Learning Radiomic Nomogram to Predict Lymph Node Metastasis for Gastric Cancer

Abstract

Background: To build and validate a spectral CT-based deep learning radiomic model for preoperative lymph node metastasis (LNM) and prognosis prediction in gastric cancer. Methods: A total of 204 pathologically confirmed gastric adenocarcinoma patients were retrospectively enrolled, and divided into training set (n=136) and test set (n=68). All of them underwent spectral CT scans before surgery. Radiomic features, containing deep learning features and hand-crafted features, were extracted from biphasic (arterial phase, AP; venous phase, VP) and three energy (40, 65, 100 keV) enhanced images. Clinical information, CT parameters and follow-up data were also collected. A radiomic nomogram for LNM prediction was built using machine learning method and evaluated in test set. Its prognostic performance was determined with Harrell's concordance index (C-index) based on patients' clinical outcomes. Findings: The biphasic, triple energy image-based radiomic signatures were associated with LNM in the two sets (P <0.001), and demonstrated good performance for discriminating LNM in test set with AUCs of 0.711 (AP CT signature; 95% CI: 0.585-0.838) and 0.755 (VP CT signature; 95% CI: 0.638-0.873). The nomogram that incorporated two radiomic signatures and CT-reported lymph node status exhibited high diagnostic ability for LNM (P <0.001), with AUCs of 0.839 (0.773-0.904) in training set and 0.821 (0.722-0.920) in test set. Moreover, the nomogram exerted a promising prognostic ability with C-indices of 0.637 (0.544-0.730; P=0.0038) for progression free survival and 0.669 (0.560-0.778; P=0.0023) for overall survival. Interpretation: This study provided a new insight into the role of a spectral CT-based radiomic nomogram for LNM and prognosis prediction of gastric cancer. Funding Statement: National Natural Science Foundation of China (81271573, 81372370, 81771924, 81501616, 81227901), National Key R&D Program of China (2017YFC1308700, 2017YFA0205200, 2017YFC1309100), the Beijing Natural Science Foundation (L182061), the Bureau of International Cooperation of Chinese Academy of Sciences (173211KYSB20160053), and the Youth Innovation Promotion Association CAS (2017175). Declaration of Interests: : The authors has nothing to disclose any potential conflicts (financial, professional, or personal) that are relevant to the manuscript. Ethics Approval Statement: Institutional Review Board approval was obtained. Written informed consent was waived by the Institutional Review Board of Zhengzhou University, because this is a retrospective diagnostic study.