Automated Recognition of Human Structure from Torso CT Images

X. Zhou, N. Kamiya, T. Hara, H. Fujita, R. Yokoyama, T. Kiryu, and H. Hoshi (Japan)

Keywords

CT images, image processing, and computeraided diagnosis

Abstract

We have been developing a computer-aided diagnosis (CAD) scheme for automatically recognizing human structure that is constructed by skin, fat, muscle, bone and internal organ regions from high-resolution torso CT images. We show some initial results for extracting skin, fat, muscle and skeleton regions. 139 patient cases of torso CT images (male: 92, female: 47, age: 12-88) were used in this study. Each case was imaged with a common protocol (120kV/320mA) and covered the whole torso with 0.63 (mm) isotopic spatial resolution and 12 (bits) density resolution. A gray-level thresholding based procedure was applied to separate the human body from background. The density and distance features to body surface were used to determine skin, and separate fat and muscle from the other organ regions. A 3-D region growing based method was used to extract skeleton. We applied this system to 139 cases and found that the skin, fat, muscle and skeleton regions were recognized correctly for 93% patient cases. The accuracy of segmentation results was acceptable by evaluating the results slice by slice. This scheme will be included in a CAD system for detecting and diagnosing the abnormal lesions in multi-slice torso CT images.

Important Links:



Go Back