Regionala ortogonala moment för datoriserad analys av texturen i mikroskopibilder
Tidsperiod: 2015-01-01 till 2018-12-31
Projektledare: Ida-Maria Sintorn
Medarbetare: Carolina Wählby, Sven Nelander
Budget: 3 643 000 SEK
Precise and quantitative image feature descriptors hold the potential to discern/detect subtle morphological differences hardly discernable by application experts. This, in combination with the increasing amounts of image data produced by today?s microscopes motivates the development of automated image analysis methods. We will investigate and characterize a novel approach for texture analysis, regional orthogonal moments (ROMs). The idea, recently proposed by the main applicant, is to utilize the descriptive strength of orthogonal moments in a filtering fashion, i.e., creating ROM filter banks. We systematically characterize the properties of ROM filter banks with different bases. We will explore and design quantitative descriptors from the filter bank responses evaluated with special consideration to noise, rotation, contrast, scale, and generalization performance; important factors in applications with natural images. In order to do this we will utilize and expand available image texture datasets and adapt machine learning methods for microscopy image prerequisites.To validate our ROM texture analysis framework, we will apply it in two clinical research projects where visual image analysis is considered infeasible or too inexact: viral pathogen detection and identification in miniTEM images, Glioblastoma phenotyping for disease modeling and personalized treatment.