The recent development of medical devices enables us to acquire high resolution medical images. 3D shape models of human organs obtained from the images are useful for computer aided systems for therapy and diagnosis. On the contrary, Artificial Intelligence（AI）-based methods and technologies have achieved high performance in medical applications. In this paper, we introduce computer medical support systems using 3D organ models and AI presenting our latest research results.

We have proposed an approach to evaluate delineation accuracy of gross tumor volume (GTV) contours with considering interobserver variability in reference contours with Dice's similarity coefficient (DSC) and Hausdorff distance (HD). The DSC and HD denote the degree of region similarity between a reference region and a compared region. Impacts of metal artifact reduction (MAR) technique on radiation treatment planning (RTP) including the delineation of GTV without metal artifacts contours considering the reference variability of 6 radiation oncologists in RTP have been investigated by using the proposed approach. Planning computed tomography (CT) images without and with metal artifacts were acquired by scanning an anthropomorphic skull phantom including a simulated GTV without and with putting crowns on teeth, respectively. The planning image without metal artifacts was considered a reference image. Six radiation oncologists delineated three contours of the GTV on CT images without and with MAR (w/oMAR and wMAR), and the reference image. The average DSC and HD between the reference and wMAR images were 0.62 and 4.43 mm, respectively,whereas they were 0.56 and 4.85 mm, respectively, between the reference and w/oMAR images (P < 0.05). In addition,there were no statistically significant differences between the reference and wMAR images with respect to D90 and D95,which were defined as the minimum doses delivered to 90% and 95% of planning target volumes (PTVs), respectively (P > 0.05). The proposed approach could evaluate the delineation accuracy of GTV contours with taking into account interobserver variability in reference contours by showing the impact of the MAR technique on the RTP.

The image rotation processing in the digital X-ray image is useful in performing the rotation processing of less than 90 degrees, however, the influence on the image quality has not been clarified. The aim of this research is to evaluate the influence of image rotation processing on image quality of digital X-ray image using physical evaluation and image evaluation. An original digital X-ray image was taken at 0 degrees. Images were taken after increasing the inclination angle at 5 degree intervals, up to 45 degrees. We measured the processed modulation transfer functions（processed MTFs）and normalized noise power spectrum（NNPS）as the physical evaluation. We measured Contrast-Detail（C-D）curve and inverse of Image Quality Figure（IQFinv）using CDRAD phantom as the image evaluation. Compared with the original image, the processed MTFs increased as the rotation angle increased in the arbitrarily rotated image, and the NNPS increased. The C D curve was not influenced by the extent of rotation, and it made no difference to IQFinv（p > 0.05）either. The result showed that the effects of the image rotation on clinical X - ray image differs depending on the structure of radiographed objects.