Purpose: The radiologists often inspect several MR images before they give a conclusion of a liver disease. However, some of the cases are falsely categorized because of the complicated relationship and subtle difference among these images. Our purpose is to establish a computer-aided diagnosis (CAD) system for distinguishing the pathologies of focal liver lesions in MR images, which helps radiologists to integrate the imaging findings with different pulse sequences and raise the diagnostic accuracy even with radiologists inexperienced in liver MR imaging.
Methods and Materials: By using 320 MR images from 80 patients with liver lesions confirmed by an experienced radiologist, we developed a software named "LiverANN" based on artificial neural network (ANN) technology. The structure of ANN was a conventional three-layer feed-forward neural network with 5 input units, 8 hidden units and 5 output units, trained with the well known back-propagation (BP) algorithm. 50 cases were selected to train the ANN. In each patient, regions of focal liver lesion on T1-weighted, T2-weighted, and gadolinium-enhanced dynamic MR images obtained in the hepatic arterial and equilibrium phases were placed by a radiologist, then the program automatically calculated the brightness and homogeneity into numerical data within the selected areas as the input signals to the ANN. As the teacher signals of the ANN, The outputs from the ANN were the 5 categories of focal hepatic diseases: liver cyst, cavernous hemangioma, dysplasia, hepatocellular carcinoma, and metastasis. The other 30 cases were used for testing the performance of the ANN. The LiverANN was built by RealBasic3.0 language, which programmed on Macintosh and could run on either Mac OS or Window OS. This would make the users convenience of using the software by their selections. With the sendmail function of LiverANN, radiologists in different places could easily send unknown data to us just simply input the type of the lesion it should be, this would be one of important sources of the retrained data of ANN.
Results: The initial experimental result showed that the LiverANN classified 5 types of liver lesions with a training accuracy of 100% on the 50 cases in the training set and a testing sensitivity of 93.3%(28/30) for the 30 test cases.
Conclusion: The experiment demonstrated the ability of ANN to fuse the complex relationships among the imaging findings with different sequences, and the ANN-based software may provide radiologists with referential opinion during the radiologic diagnostic procedure.
Questions about this event email: firstname.lastname@example.org