RESUMO
BACKGROUND: Lymph node metastasis (LNM) in patients with intrahepatic cholangiocarcinoma (iCCA) affects treatment strategies and prognosis. However, preoperative imaging is not reliable enough for identifying LNM. PURPOSE: To develop and validate a radiomics nomogram based on dynamic contrast enhanced (DCE)-MR images for identifying LNM and prognosis in iCCA. STUDY TYPE: Retrospective. SUBJECTS: Two hundred four patients with pathologically proven iCCA who underwent curative-intent resection and lymphadenectomy (training cohort: N = 107, internal test cohort: N = 46, and external test cohort: N = 51). FIELD STRENGTH/SEQUENCE: T1- and T2-weighted imaging, diffusion-weighted imaging and DCE imaging at 1.5 T or 3.0 T. ASSESSMENT: Radiomics features were extracted from intra- and peri-tumoral regions on preoperative DCE-MR images. Imaging features were evaluated by three radiologists, and significant variables in univariable and multivariable regression analysis were included in clinical model. The best-performing radiomics signature and clinical characteristics (intrahepatic duct dilatation, MRI-reported LNM) were combined to build a nomogram. Patients were divided into high-risk and low-risk groups based on their nomogram scores (cutoff = 0.341). Patients were followed up for 1-102 months (median 12) after surgery, the overall survival (OS) and recurrence-free survival (RFS) were calculated. STATISTICAL TESTS: Receiver operating characteristic (ROC) curve, calibration, decision curve, Delong test, Kaplan-Meier curves, log rank test. Two tailed P < 0.05 was considered statistically significant. RESULTS: The nomogram incorporating intra- and peri-tumoral radiomics features, intrahepatic duct dilatation and MRI-reported LNM obtained the best discrimination for LNM, with areas under the ROC curves of 0.946, 0.913, and 0.859 in the training, internal, and external test cohorts. In the entire cohort, high-risk patients had significantly lower RFS and OS than low-risk patients. High-risk of LNM was an independent factor of unfavorable OS and RFS. DATA CONCLUSION: The nomogram integrating intra- and peri-tumoral radiomics signatures has potential to identify LNM and prognosis in iCCA. EVIDENCE LEVEL: 3 TECHNICAL EFFICACY: Stage 2.
RESUMO
RATIONALE AND OBJECTIVES: To develop an efficient machine-learning model using pituitary MRI radiomics and clinical data to differentiate growth hormone deficiency (GHD) from idiopathic short stature (ISS), making the diagnostic process more acceptable to patients and their families. MATERIALS AND METHODS: A retrospective cohort of 297 GHD and 300 ISS children (4-12 years) were enrolled as training and validation cohorts (8:2 ratio). An external cohort from another institution (49 GHD and 51 ISS) was employed as the testing cohort. Radiomics features extracted from the anterior pituitary gland on sagittal T1-weighted image (1.5 T or 3.0 T) were used to develop a radiomics model after feature selection. Hematological biomarkers were selected to create a clinical model and combine with the optimal radiomics features to create a clinical-radiomics model. The area under the receive operating characteristic curve (AUC) and Delong test compared the diagnostic performance of the previously mentioned three models across different validation and testing cohorts. RESULTS: 17 radiomics features were selected for the radiomics model, and total protein, total cholesterol, free triiodothyronine, and triglyceride were utilized for the clinical model. In the training and validation cohorts, the diagnostic performance of the clinical-radiomics model (AUC=0.820 and 0.801) was comparable to the radiomics model (AUC=0.812 and 0.779, both P >0.05), both outperforming the clinical model (AUC=0.575 and 0.593, P <0.001). In the testing cohort, the clinical-radiomics model exhibited the highest AUC of 0.762 than the clinical and radiomics model (AUC=0.604 and 0.741, respectively, P <0.05). In addition, the clinical and radiomics models demonstrated similar diagnostic performance in the testing cohort (P >0.05). CONCLUSION: Integrating radiomics features from conventional pituitary MRI with clinical indicators offers a minimally invasive approach for identifying GHD and shows robustness in a multicenter setting.