Tumor texture analysis in 18F-FDG PET/CT.
18F-FDG PET is often used in clinic for cancer diagnosis, staging and therapy planning and monitoring. Alternative parameters to the classical semi-quantitative features have been recently proposed to study the heterogeneity of tumors though texture analysis and quantify uptake distribution among them. However, too many texture features are available so the aim of the current project is reduce the set of possible variables to avoid computational cost and data redundancy.
A retrospective analysis of 53 head and neck and 12 rectal cancer patients was performed. The images were acquired in different PET devices following different protocols and post-processing and reconstruction techniques. A workflow in an open-source software, 3D slicer, was designed for the project. Expert clinicians were trained to use the software and the selected modules for tumor segmentation. After tumor segmentation 66 features were calculated and a database was created for each cancer type. Statistical analysis and dimensionality reduction techniques were performed on the data.
Dimensions were reduced to five and three independent factors or components for head and neck and rectal cancer cohort, respectively, and an approximation of which were the possible original features more representative for the data was done. Future work must be focused on building a discriminative classifies based on those factors that could be predictive o response.