[Methods and feasibilities of texture analysis in PET diagnostics]

One of the current research objectives of medical imaging is to determine the prognostic value of tumor textures and related numerical values. In PET/CT studies the diagnostic and prognostic values of specific texture parameters were confirmed at several tumor types (lung, prostate, cervix, colon, head and neck). However, the results are often contradictory, various publications find different texture parameters useful for the same tumor type. The reason for the contradictions is partly methodological, since the definition and the calculation of texture data is a multi-step process. Such steps include scan protocol, image reconstruction, tumor segmentation, re-sampling the voxel values and the form of texture algorithms. Recent publications show that by harmonizing these steps, the prognostic power and reliability of the texture features can be improved. The most optimal way of harmonization would be a special phantom application that could simulate inhomogeneous distributions typical for tumor tissues, with high reproducibility.