Affiliations: [a]
Department of Radiology, the First Affiliated Hospital of Nanjing Medical University, Nanjing, China
| [b]
Department of Obstetrics & Gynecology, the First Affiliated Hospital of Nanjing Medical University, Nanjing, China
Correspondence:
[*]
Corresponding authors. Ting Chen, PhD and Jiacheng Song, MD., Department of Radiology, The First Affiliated Hospital of Nanjing Medical University, 300 Guangzhou Road, Nanjing 210029, China. Tel.: +86 25 68136861; Fax: +86 25 83724440; E-mail: [email protected]. (T.C.), E-mail: [email protected]. (J.S.)
Note: [1] These authors contributed equally to the study as the first author.
Abstract: OBJECTIVE:To investigate the value of nomogram analysis based on conventional features and radiomics features of computed tomography (CT) venous phase to differentiate metastatic ovarian tumors (MOTs) from epithelial ovarian tumors (EOTs). METHODS:A dataset involving 286 patients pathologically confirmed with EOTs (training cohort: 133 cases, validation cohort: 68 cases) and MOTs (training cohort: 54 cases, validation cohort: 31 cases) is assembled in this study. Radiomics features are extracted from the venous phase of CT images. Logistic regression is employed to build models based on conventional features (model 1), radiomics features (model 2), and the combination of model 1 and model 2 (model 3). Diagnostic performance is assessed and compared. Additionally, a nomogram is plotted for model 3, and decision curve analysis is applied for clinical use. RESULTS:Age, abdominal metastasis, para-aortic lymph node metastasis, location, and septation are chosen to build Model 1. Ten optimal radiomics features are ultimately selected and radiomics score (rad-score) is calculated to build Model 2. Nomogram score is calculated to build model 3 that shows optimal diagnostic performance in both the training (AUC = 0.952) and validation cohorts (AUC = 0.720), followed by model 1 (AUC = 0.872 for training cohort and AUC = 0.709 for validation cohort) and model 2 (AUC = 0.833 for training cohort and AUC = 0.620 for validation cohort). Additionally, Model 3 achieves accuracy, sensitivity, and specificity of 0.893, 0.880, and 0.926 in the training cohort and 0.737, 0.853, and 0.613 in the validation cohort. CONCLUSION:Model 3 demonstrates the best diagnostic performance for preoperative differentiation of MOTs from EOTs. Thus, nomogram analysis based on Model 3 may be used as a biomarker to differentiate MOTs from EOTs.
