Original Jpn J Radiol Technol 2000 ;
Development of an Image-understanding System for Double-contrast Gastric Radiograms

SHIGEHIRO FUKUSHIMA and KOUICHI YOSHIMOTO1)
Graduate School of Design, Kyushu Institute of Design
1)Faculty of Computer Science and Systems, Engineering, Kyushu Institute of Technology
(Present address: Ikuhashi City Hall)
Received Nov. 1, 1999; Revision accepted Jan. 11, 2000; Code No. 532
Summary
An image-understanding subsystem was developed for the first time for the computer-aided diagnosis of double-contrast gastric radiograms, the most effective images for diagnosing gastric cancer. Three tasks were incorporated: the gastric region was recognized by line tracking or by using active contours; barium-pool regions were recognized by region extraction; and the spinal region was recognized by examination of gray-level profiles. The image-understanding performance was evaluated by application to 18 direct radiograms. On the severe side, the gastric region was very well recognized up to 83% by the line-tracking method and up to 89% by the active-contour method. Barium-pool regions were well recognized up to 78% and the spinal region up to 56%. On the more optimistic side, the gastric region and barium-pool regions were recognized up to 100%, and the spinal region was recognized up to 83%. Furthermore, the computation time was evaluated for each processing element. The quality of image understanding by the developed system is also described.
Key words: Computer-aided diagnosis, Image understanding, System development, Labeling, Evaluation