Identification of Biliary Atresia in Infantile Cholestasis: Integrating Radiomics With MRCP for Unobservable...

Magnetic resonance cholangiopancreatography (MRCP) may assist in the workup of infantile cholestasis as nonvisualization of the biliary tree is seen with biliary atresia (BA). However, this finding can also be seen with other causes of infantile cholestasis. The purpose of this study is to differentiate BA from other causes of infantile cholestasis using a classification tool integrating MRCP-based radiomics and clinical signatures in patients with nonvisualization of the extrahepatic biliary tree on MRCP.

Data from infants with cholestasis due to BA, cytomegalovirus infection, or idiopathic neonatal hepatitis (INH) from 2 sites was collected. Radiomics features from MRCP images were selected using Spearman and LASSO methods, followed by applying the optimal machine learning model to develop a radiomics signature. Clinical factors showing significant differences between BA and non-BA groups in training cohort were used to develop a clinical signature using the model. A nomogram model incorporating the signatures was developed. The nomogram model and signatures' performance were assessed using the area under the curve (AUC), accuracy, sensitivity, specificity, precision, and F1 score. The DeLong test, decision curve analysis (DCA), calibration curves, and the Hosmer-Lemeshow test were utilized to evaluate the nomogram model.

The training cohort consisted of 112 cases (62 BA and 50 non-BA) from site 1, while the external validation cohort included 35 cases (20 BA and 15 non-BA) from site 2. After screening, 2 clinical factors and 8 radiomics features were included. The signatures were fitted using the K-Nearest Neighbors model. The nomogram model showed an AUC of 0.981 in the training cohort and 0.913 in the external validation cohort, significantly outperforming both the signatures in the training cohort and the clinical signature in the external validation cohort, as confirmed by the DeLong test. The DCA indicated the clinical utility of the model. The Calibration curves and the Hosmer-Lemeshow test confirmed the model's adequate fit.

The nomogram model may hold clinical utility. In our cohorts, it was effective for identifying BA among cases with infantile cholestasis attributed to BA, cytomegalovirus infection, or INH in scenarios where the extrahepatic biliary system is not visualized on MRCP.