Development of an Ultrasound Prediction Model to Discriminate between Malignant and Benign Liver Lesions

To discriminate between malignant and benign liver lesions, we evaluated the ultrasound features of the target lesions in 266 patients and established a prediction model using a logistic regression algorithm. The prediction model based on independent factors was expressed as predictive score = 1.129 × interaction of irregular shape and unclear boundary + 1.398 × occupying effect + 2.363 × hypo-echoic halo + 1.987 × marginal vascular sign + 3.627 × cirrhosis background + 2.976 × nodule in nodule sign + 3.690 × metastasis sign. Receiver operating characteristic curve analysis revealed that the optimal cutoff predictive score was 2.8 (area under the curve = 0.942). The specificity of the prediction model was not significantly different from that of computed tomography/magnetic resonance imaging (91.7% vs. 98.8%, p = 0.077), whereas the prediction model had a lower sensitivity (90.1% vs. 97.8%, p < 0.001) and accuracy (90.6% vs. 98.1%, p < 0.001) than computed tomography/magnetic resonance imaging. We conclude that the ultrasound prediction model exhibited good diagnostic performance in discriminating malignant from benign liver lesions.