It is adaptive and excellent ability to extract features that enables artificial intelligence based on deep learning to achieve significant success. In this project, we developed a method to locally evaluate and extract the feature called Multifractal Spectrum (MFS), taking images as an example. Multifractal Spectrum was devised by Xu, Ji and Fermüller in 2009. While it is a global feature to evaluate the complexity of image textures in multilayer levels, it has a disadvantage to extract only averaged properties of images. It is the purpose of this project to make it possible to locally evaluate them, and resulted feature cannot be obtained by deep learning such as convolutional neural networks (CNN). We are now applying this new feature-extracting method to the diagnosis of duodenal tumors and of tumor grade of urothelial carcinoma and so on.