Texture analysis for identifying heterogeneity in medical images

    Heterogeneity is a well-recognized feature of malignancy associated with increased tumor aggression and treatment resistance. Texture analysis (TA) of images of various modalities, including, among others, CT, MRI or PET, can be applied to quantify the tumor heterogeneity and to extract useful information from images acquired in routine clinical practice without additional radiation or expense of further procedures. In this paper, we elaborate on the filtration-based approach to TA applied for extracting features from large sets of simulated images reflecting various clinical circumstances. The areas under receiver operating characteristic curves were used to assess the diagnostic performance of the derived biomarkers. We present and discuss their discriminative abilities in identifying heterogeneity and classifying images with simulated lesions of various characteristics and localized density variations.