Xiangrong Zhou PhD, Presenter: Nothing to Disclose

Syoichi Morita, Abstract Co-Author: Nothing to Disclose

Takeshi Hara PhD, Abstract Co-Author: Nothing to Disclose

Ryujiro Yokoyama, Abstract Co-Author: Nothing to Disclose

Huayue Chen, Abstract Co-Author: Nothing to Disclose

Masayuki Kanematsu MD, Abstract Co-Author: Nothing to Disclose

Hiroshi Fujita PhD, Abstract Co-Author: Nothing to Disclose

Understanding anatomical structures on CT images is an essential step of computer-based medical image analysis. Partitioning whole body region on a CT scan into a number of special organ or tissue units based on anatomical definition is the way to realize this tough task. Here, an organ unit is generally presented by a rotated bounding box that tightly covers a special organ region with a little background as small as possible. In this work, we demonstrate a computer system that can automatically and quickly partitioning a 3D volumetric CT scan into a number of organ or tissue boxes based on anatomical definition.

A dataset consisting of more than 4,000 volumetric CT scans was used for performance evaluation. These CT images were generated from 4 kinds of CT scanners by using different scan protocols for clinical diagnosis. The spatial resolution of the CT images varied from 0.62 mm to 5 mm. Eighteen major organs and tissues including heart, liver, gallbladder, spleen, pancreas, stomach, bladder, left-lung, right-lung, left-kidney, right-kidney, left-femur-head, right-femur-head, left-psoas-major-muscle, right-psoas-major- muscle, uterus and rectum, abdominal rectus muscle, and inferior vena cava (IVC) with ventral aorta were selected as partitioning targets. In this study, the bounding box was considered to be correct if most (two thirds of the volume) of the detected 3D box and the manually inputted box overlapped with each other.

Our evaluation showed that the success rates for most targets were in the range of 95 % to 100 %, besides of stomach (94 %), spleen (90 %), pancreas (82 %) and gallbladder (84 %). Typical computing time for partationing a torso CT scan using a general-purpose computer equipped with an Intel Core2Duo 2.23-GHz CPU was 2 minutes or less.

We developed a computer system that can be used to partition a CT scan into more than 18 kinds of major organ or tissue units automatically. The efficiency and accuracy were validated using a large dataset including more than 4,000 real clinical CT scans.

http://abstract.rsna.org/uploads/2014/14010105/14010105_cq23.jpg

Zhou, X, Morita, S, Hara, T, Yokoyama, R, Chen, H, Kanematsu, M, Fujita, H, Automatic Partitioning Torso CT Images based on Anatomical Definition and its Performance Evaluation by using a Large Dataset.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14010105.html