Recently, we often deal with multi-dimensional image data such as 3D volume data and video data. Especially, in the medical image processing field, we are required to process 3D volume data, for instance CT, MRI and PET processing. In general, however, it is not easy to construct complex 3D processing procedures manually compared with 2D ones. Therefore, a lot of knowledge and experiences about image processing are necessary in this field. A lot of multi-dimensional image processing algorithms which can apply complex problem have been developed. However most of these methods require adjusting many parameters and processing procedures manually by trial and error every other problem. Thus, we propose a new method named "3D-ACTIT; Automatic Construction of Tree-structural Image Transformation" to make various 3D image processing procedures automatically. 3D-ACTIT constructs tree-structured image processing filter with one input one output filter and two input one output filter by using Genetic Programming to satisfy the given several image examples. In this paper, we apply 3D-ACTIT to 3D-PET image data and construct the processing which extracts kidney and lung regions. Experimental results show that complex 3D image processing procedures can be constructed automatically.
抄録全体を表示