Three-dimensional feature understanding based on curvature information of FDG-PET
Sota Takamuro, Tetsuya Tozaki, Michio Senda · 2018
The cancer diagnosis based on FDG-PET has become popular in recent years, however false positive shadows occur due to the characteristics of the imaging method. Curvature has invariant properties in analyzing direction and position of organ shape. In this research, we focused on curvature information of PET image. We obtained curvature by getting the Hessian matrix in each voxel of the three-dimensional PET image and calculating its eigenvalues. We confirmed that image based on curvature intensity shows emphasis of liner shadow and outline of tissues. In addition, we expressed running direction of tissues by changing each component of curvature vector into color. Analyzing combination of principal curvature showed certain feature of isolated shadows. Moreover, we expressed more detail running direction on isolated shadows. This paper shows the detailed results of applying these methods to rectal cancer.