[Effect of a comprehensive image processing on radiologists' performance in differential diagnosis of liver lesions using CT and MRI].

PURPOSE

To evaluate whether a comprehensive image processing method as CAD using CT and MRI can improve the radiologists' diagnosis performance in the differentiation of focal liver lesions.

METHOD AND MATERIALS

A clinical image database used in this study consists of 14 cases of each lesion including hepatic cysts, hepatocellular carcinoma (HCC), metastatic liver cancer, and hemangioma. This technique by using MR images obtained with various imaging sequences and a series of dynamic MR and dynamic CT images is designed for the enhancement of liver lesions pixel by pixel. In this method, we make the pixel sizes of MR images the same size of CT image by using tri-linear interpolation technique. Then the 3D image registration technique based on mutual information is applied for the matching of images. The image intensity pattern with and without contrast enhancement is determined as the template for the differential detection of each lesion. Pixel-by-pixel cross-correlation coefficient is calculated for the enhancement of each lesion. The radiologists' performance in distinguishing between the liver lesion was evaluated by receiver operating characteristic analysis (ROC) with a continuous rating scale.

RESULTS

In free-response ROC analysis, true positive fractions were 75%, 87%, 85%, and 86% for hepatic cysts, HCC, metastatic liver cancer and hemangioma, respectively. Furthermore, average number of false positive and false negatives per image was 3.4 and 0.3, respectively. When radiologists made differential diagnosis of the liver lesions with the images of this technique, diagnostic accuracy was statistically significantly improved compared to the diagnostic accuracy without the images of this technique. The average area under the ROC curve (Az value) improved from 0.881 to 0.964 (p=0.069) for the differential diagnosis of hepatic cysts. Furthermore, the Az value of HCC, metastatic liver cancer, and hemangioma improved from 0.951 to 0.979 (p=0.040), from 0.946 to 0.976 (p=0.226), and from 0.966 to 0.987(p=0.045), respectively.

CONCLUSION

A comprehensive image processing method as CAD using CT and MRI can improve the radiologists' diagnostic performance in the differentiation of focal liver lesions. CLINICAL RELEVANCE/APPLICATION: This method improved the performance of differential detection of liver lesions from a large number of images and it would save radiologists' reading time, and thus could assist their diagnosis.