Discrimination of myelofibrosis and liver cirrhosis using a spleen ultrasound-based radiomics nomogram model

The aim of the present study was to develop and validate a spleen ultrasound-based radiomics nomogram for non-invasive differentiation between myelofibrosis (MF) and liver cirrhosis. A total of 90 patients (40 patients with MF and 50 with cirrhosis) were enrolled, and randomly divided into training (n=72) and validation (n=18) sets. Ultrasound radiomics features were extracted from spleen images using PyRadiomics, and feature selection was performed using the least absolute shrinkage and selection operator and Boruta algorithms. Multivariate logistic regression combining radiomics features and clinical parameters was used to construct the nomogram model. Two radiomics features (wavelet.HHL_glcm_ClusterShade and original_glszm_GrayLevelNonUniformity) were selected. When combined with direct bilirubin and glutamic-oxalacetic transaminase levels, the nomogram achieved an area under the curve of 0.958 [95% confidence interval (CI), 0.912-1.0] in the training set and 0.750 (95% CI, 0.487-1.0) in the validation set, with sensitivities of 93.8 and 100%, specificities of 87.5 and 60%, and accuracies of 80.3 and 71.4%, respectively. In conclusion, the spleen ultrasound-based radiomics nomogram demonstrated promising performance for differentiating MF from liver cirrhosis, providing a potential non-invasive tool for early diagnosis and clinical decision-making.