Abstract
Model for End-stage Liver Disease (MELD) is a common score used in clinical practice to estimate the prognostic outcome of cirrhotic patients. This score is obtained from laboratory results.
Here a novel method is proposed to estimate the MELD score based on textural information extracted from normalized ultrasound (US) images of liver parenchyma. The information obtained from the co-ocorrence matrix and the monogenic decomposition of the image is linearly combined to compute the score. The application of US data for prognosis purposes is also a noteworthy novelty of this paper.
A dataset of 82 cirrhotic patients is used in this work. An optimal cut-off from a ROC analysis lead to an accuracy of 80% and an AU-ROC of 0.801 in the prognosis prediction. No statistical differences were found between the MELD score and the proposed US score and a strong correlation (0.65 p < 0.01) was attained.