Radiomics-based diagnosis of acute cholangitis using non-contrast computed tomography: a preliminary study

We developed a radiomics model for the diagnosis of acute cholangitis using non-contrast computed tomography (NCCT).

We included 79 patients with acute cholangitis and 79 without known liver disease; all patients underwent NCCT and contrast-enhanced CT (CECT). Patients were randomly divided into a development dataset (n = 118) and a test dataset (n = 40). In total, 999 radiomics features were extracted from each segmented liver. Two experienced radiologists interpreted the NCCT for evidence of acute cholangitis by a consensus-based approach. Subsequently, both radiologists re-evaluated each case using the arterial and delayed phase CECT. Diagnostic performance was assessed using sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUROC) for the NCCT-based radiomics model, NCCT-based radiologists' evaluation, and CECT-based radiologists' evaluation. McNemar's test was used to compare sensitivities and specificities.

In the test dataset, the NCCT-based radiomics model demonstrated an AUC-ROC of 0.888 for diagnosing acute cholangitis. Sensitivity, specificity, and accuracy were 95%, 75%, and 85%, respectively. Radiologists' evaluation achieved a sensitivity of 65%, specificity of 100%, and accuracy of 82.5% when using NCCT. These metrics improved to 95% sensitivity, 100% specificity, and 97.5% accuracy when re-evaluated with additional CECT images. The performance of the NCCT-based radiomics model was comparable to that of radiologists' evaluation using CECT images (p = 0.68).

The NCCT-based radiomics models demonstrated potential in diagnosing acute cholangitis; this approach may offer a viable alternative to CECT, particularly in patients with contraindications to contrast agents.