Method for Segmentation of Structures in the Serially Sectioned Images of the Entire Body and Surface Reconstruction to Make Three-Dimensional Images

Abstract

"Propose : First purpose of this research is to prepare the segmented images of whole body in detail, which are distributed to help other investigators make 3D images and virtual dissection software of whole body. Second purpose is to present of semi-automatic surface reconstruction on the commercial software, which enables any researchers to do surface reconstruction quickly and objectively for making 3D images without the help of computer engineers.

Materials & Methods : Three-hundred nineteen structures (114 left lower limb structures, 105 head and neck structures, 15 heart structures, 85 left upper limb structures) of whole body were decided to segment in 1,702 temporary segmented images (PSD file, interval 1.0 mm, resolution 2,468 X 1,407) including anatomical images. On the PhotoshopTM, selections which fit the structures' contours were drawn automatically, semi-automatically, or manually; subsequently, the selections were put into the layers. After filling the selections with colors, the temporary segmented images were converted to segmented images (TIFF files). The segmented images were staked to make coronal and sagittal segmented images for verifying segmentation. One-hundred fourteen structures (1 skin, 32 bones, 7 knee joint structures, 60 muscles, 7 arteries, 7 nerves) in left lower limb were decided to semi-automatic surface reconstruction on the commercial software. Steps of semi-automatic surface reconstruction were as follows. First, disassembled segmented images were made by disassembling selections. Second, contour images were made by stacking selections. Third, contour 3D images were made by filling contour gaps with surfaces. Fourth, 3D images were made by deleting contours. Fifth, assembled 3D images were made by assembling 3D images.

Results : Contours of the anatomical structures in the three-hundred nineteen structures segmented images were corrected, and verified by examining the coronal images, sagittal images, and browsing software of Visible Korean Human. On the Maya, the 3D images of selected anatomic structures in left lower limb could be displayed opaquely or semi-transparently and rotated, which revealed that the 3D images were fit for anatomic knowledge.

Conclusion : The segmentation techniques of this research can be used to segment many structures in other images quickly and correctly. These technical modifications of existing software will provide new solutions in medical education and research. Also, in this research, method of semi-automatic surface reconstruction on commercial software was developed. The method could enable other researchers to do surface reconstruction quickly and objectively for making 3D images."