10027/22974 Eleonora D'Arnese Eleonora D'Arnese Automating Lung Cancer Identification in PET/CT Imaging University of Illinois at Chicago 2018 Lung cancer identification Image processing FDG-PET/CT 2018-11-27 00:00:00 Thesis https://indigo.uic.edu/articles/thesis/Automating_Lung_Cancer_Identification_in_PET_CT_Imaging/10783967 Early and accurate diagnosis of lung cancer is one of the most investigated open challenges in the last decades. The diagnosis for this cancer type is usually lethal if not detected in early stages. For these reasons it is clear the need of creating an automated diagnostic tool that requires less time for the identification and do not require a cross validation of the results by different radiologists, being in this way cheaper and less error prone. The aim of this work is to implement a completely automated pipeline that starting from the current imaging technologies, such as Computed Tomography (CT) and Positron Emission Tomography (PET), it will identify lung cancers to be employed for the staging; moreover, it will be a suitable starting point for a machine learning based classification procedure. In particular, this project proposes both a methodology and the related software tool that taking as input Digital Imaging and COmmunications in Medicine (DICOM) files of chest PET and CT and by exploiting the characteristic of both of them is capable of automatically identify the lungs and the eventually presence of tumor lesions. Going deeper into the topic and analyzing the literature it is possible to notice as very different solutions have been proposed in literature for an accurate and fast identification of lung cancer. Often such solutions start from CT and PT but generally they are semiautomatic tools that still require the intervention of a physician in charge of indicating which is the Region of Interest (ROI). On the other side, when fully automatic approaches has been proposed, they are closed source with not available datasets, so it is pretty impossible a comparison, as first, and to use their results to move forward machine learning based approaches. For this reason, this work proposes a methodology and its technical implementation that aim at being a reference point for future work into the field. As it will be possible to see, the thesis main contribution is about the proposition of a fully automated identification of the Region of Interest in the medical images, a fully automated segmentation procedure able to find lung cancer lesion inside the ROI and a technique for combine information obtained from both CT and PET. A validation of the pipeline will be also discussed, measuring both the execution time and the obtained accuracy. Moreover, some consideration about future developments of this project will be proposed.