Πλοήγηση ανά Συγγραφέας "Skepasianos, Iraklis"

Τώρα δείχνει 1 - 1 of 1
  • Μικρογραφία εικόνας
    Τεκμήριο
    Texture kinetics and multiscale texture analysis for predicting breast cancer treatment response.
    (ΕΛ.ΜΕ.ΠΑ., ΣΧΟΛΗ ΜΗΧΑΝΙΚΩΝ (ΣΜΗΧ), Τμήμα Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών, 2020-03-31) Skepasianos, Iraklis; Σκεπασιανός, Ηρακλής
    Evaluation of tumor response has been extensively investigated using a wide variety of manual and computer assisted methods. Oncologists are using the Response evaluation criteria in solid tumors (RECIST) and World Health Organization (WHO) criteria, among others, to examine the response of the tumor in the therapy process. However, both of these criteria use the dimensions of the tumor as a feature e.g. its diameter. As a rseult, these approaches are unable to capture the heterogeneity of the tumor tissue structure, which might change after the therapy. For this reason, more accurate quantitative methods for assessing tumor response after therapy have been introduced. In this work we propose describing breast cancer tissue using texture kinetics and multi-scale texture for the prediction of neoadjuvant chemotherapy response of the patients. As aforementioned, texture features can provide information of the tumor tissue structure in order to overcome current limitations in the RECIST and WHO criteria used in clinical practice. In addition, we propose a framework of Gabor multi-scale filtering to examine the capabilities of multi-scale texture features since texture in different scale provides important information which would be not available in a single scale. Using a public dataset which includes Dynamic Contrast Enhanced (DCE) Magnetic Resonance Imaging (MRI) data, we examined the texture kinetics and multi-scale textural features, since DCE - MRI provides 3D spatiotemporal evolution of the tumor. More specifically, Radiomic features were extracted, offering a plethora of features describing the tissue heterogeneity of the tumor. Considering that, Radiomic features were examined for analyzing their predictive strength in Neo-adjuvant therapy (NAC) response. Along with this contribution, Radiomic features extracted from Gabor multi-scale filtered images were also extracted in order to address a second research question regarding the role of the scale and orientation of image texture in predicting the therapy outcome. Results showed that texture kinetics are able to improve the predictability of tumor response to NAC with an area under receiver operating characteristic curve (AUROC) sensitivity of ≈ 81%. Similarly, Gabor multi-scale texture features, provided an average accuracy of >= 70%, while the best accuracy was 88% with scale set at 0.5, confirming that texture at different scales and orientations adds value in the therapy predictive modelling.