Quantitative ultrasonic detection and classification of diffuse liver disease: Comparison with human observer performance

A multiparameter ultrasonic tissue characterization system has been developed and tested on several types of diffuse liver disease. The four tissue characterization parameters used are based on the first and second order statistics of the B-scan image. Performance of the system was evaluated using receiver operating characteristic (ROC) analysis and was compared with the performance of experienced human observers viewing B-scan images. The machine-based multiparameter system achieved an area under the ROC curve (A(z)) of 0.88 for detection of chronic hepatitis in more than 100 proven cases of the disease. This was dramatically better than the performance of human observers (A(z) = .64, P < .05) and compares favorably to the performance of other accepted diagnostic tests such as head CT and the PAP smear. For detection of Gaucher's disease, the A(z) for the system was .92, whereas for separating hepatitis from Gaucher's disease A(z) was .8.4. Human observers also did well at these tasks (P > .8) using organomegaly as their major criterion for diagnosing Gaucher's disease. For primary biliary cirrhosis the system A(z) was .80, for glycogen storage disease A(z) was .94. These results suggest that use of multiparameter tissue characterization can significantly increase the usefulness of ultrasound for evaluation of diffuse liver disease.