In this paper, we propose a method to detect liver cancers from dynamic X-ray CT images based on a two-dimensional histogram analysis. In the diagnosis of a liver, a doctor examines dynamic CT images. Dynamic CT images consist of three images, namely early phase, portal phase, and late phase, which are taken sequentially within a few minutes. Since the early and late phase images are important for diagnosing liver cancer, our method refers to both of them for detecting suspicious regions and eliminating false positives. At first, the proposed method extracts liver cancer candidates by applying an adaptive neighbor type difference filter from the late phase image. Most of the false positive regions are eliminated by two-dimensional histogram analysis of each region of interest. We applied the proposed method to 21 dynamic CT images. The result showed that sensitivity was 100\\% and false positives were 0.3 per case in average.

View full abstract