An automatic detection method for liver lesions using abdominal computed tomography

Treatments for liver cancer requires the information of liver such as its boundary, precise size and localization of tumors, and spatial relations among these tissues. A computer-aided diagnosis (CAD) system can to help doctors conveniently acquire the information and provide valuable second opinions. In our study, we aim to developing a fully automatic method for detecting liver tumors using abdominal CT images. The proposed method consists of three modules. First, a DICOM image was read and preprocessed using an adaptive liver window to enhance its contrast Then we used statistical and morphological features to extract the liver mass. Finally, we extracted texture features for each pixel in the extracted liver regions and applied neural network to classify pixels and to identify whether they were belonged to normal tissues or liver lesions. In order to validate the proposed study, we have tested our method in a database from SO liver patients. We demonstrated the accuracy of the tumor segmentation method using a cross-validation protocol and three area error metrics. The performance was evaluated using TP, FP, and FN percentages, which were 71.82%, 37.83% and 28.17%.