Institutional Repository
Technical University of Crete
EN  |  EL



My Space

Combined analysis of phenotype and genotype in lung cancer using radiogenomics framework

Dovrou Aikaterini

Full record

Year 2020
Type of Item Diploma Work
Bibliographic Citation Aikaterini Dovrou, "Combined analysis of phenotype and genotype in lung cancer using radiogenomics framework", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2020
Appears in Collections


During the last years, there is an increased interest in the development of models that intend to link cancer imaging features to the tumor genetic profile (Radiogenomics), in order to contribute in the diagnosis, evaluation, treatment planning and prognosis of lung cancer. Imaging features are extracted from the medical standard-of-care images and reflect the tumor phenotype. The tumor phenotype is formed by the rearrangement and the alterations of the genetic information. The gene mutations lead to cell proliferation and thus to cancer spread, which defines the cancer stage. There is an emerging need of valid diagnosis tools for lung cancer staging in order to define the proper treatment planning. This study aims at investigating correlations among the most significant imaging features and genes in lung cancer and their potential to detect the stage of the patients with Non-Small Cell Lung Cancer (NSCLC). The proposed analysis includes the identification of the differentially expressed genes between cancer and healthy population by the application of the Significance Analysis of Microarrays (SAM) algorithm and the 2-fold change technique. Subsequently, correlation of these genes with the Computed Tomography (CT) imaging-derived features was conducted through the Spearman rank correlation test, SAM for quantitative problems and False Discovery Rate (FDR) methods, revealing 78 significant genes correlated to imaging features. These genes were validated for their diagnostic character through classification and clustering techniques followed by the formation of clusters of co-expressed imaging features (metafeatures). From these two procedures, 77 homogeneous metafeatures and 73 significant genes were identified. These genes were analyzed with least absolute shrinkage and selection operation (LASSO) regression for their ability to predict the metafeatures accurately. Through the analysis, 51 metafeatures that are correlated and can be predicted with the genes, were identified. The last step was comprised of the examination of the predictive ability of the remaining significant genes and metafeatures in lung cancer staging through various classification tests using linear Support Vector Machines (SVM) algorithms. This study concluded that staging cancer could be predicted from a) genes, with an accuracy of 75.00% - 94.11%, b) metafeatures, with an accuracy of 70.83% - 95.00% and c) the combination of metafeatures and genes, with an accuracy of 85.24% - 100.00%. Additionally, artificial imaging features were produced from the linear combination of the genes that could replace the actual metafeatures and predict cancer staging with an accuracy of 76.47% - 83.60%. Finally, signaling and metabolism pathways as well as miRNA targets were revealed during the enrichment analysis of the derived gene signatures.

Available Files