A novel H.264-supported approach for detecting and classifying hepatic lesions in computed tomographic images

Hepatocellular carcinoma (HCC) comprises a major subtype of primary liver cancer that could be diagnosed earlier using computed tomography (CT) examination. Video compression reduces the size of multi-frame data and induces blocking effect. This study aims to examine if the blocking effect alters the performances of HCC tumor detection and hepatic lesion classification in CT images. H.264 is considered in this study because it can compress 14-bit grayscale multi-frame data that is compatible with the image requirement of CT. A range of quantization parameters (QP) was determined in a phantom study. With this QP range, the trained support vector machines (SVM) based on image features were applied to test 20 HCC cases and 21 normal cases. The SVM performed significantly better than the random classification for detecting HCC in images compressed with the QP levels of interest. Images compressed with QP1 yielded the best performance. The same range of QP levels was applied to 15 lesion-bearing images. The image quality indices of these images were calculated to form the feature vectors. The clustering of these feature vectors identified at least six clusters. The association of the lesion classes with the identified clusters was found significant for all QP levels of interest. QP5 and QP9 yielded higher association than QP-3 and QP1. The findings proved a novel application of H.264 compression for enhancing the hepatic lesion detection classification in CT images.