IMAGE PROCESSING, SEGMENTATION AND MACHINE LEARNING MODELS TO CLASSIFY AND DELINEATE TUMOR VOLUMES TO SUPPORT MEDICAL DECISION

Techniques for processing and analysing images and medical data have become the main’s translational applications and researches in clinical and pre-clinical environments. The advantages of these techniques are the improvement of diagnosis accuracy and the assessment of treatment response by means of quantitative biomarkers in an efficient way. In the era of the personalized medicine, an early and efficacy prediction of therapy response in patients is still a critical issue. In radiation therapy planning, Magnetic Resonance Imaging (MRI) provides high quality detailed images and excellent soft-tissue contrast, while Computerized Tomography (CT) images provides attenuation maps and very good hard-tissue contrast. In this context, Positron Emission Tomography (PET) is a non-invasive imaging technique which has the advantage, over morphological imaging techniques, of providing functional information about the patient’s disease. In the last few years, several criteria to assess therapy response in oncological patients have been proposed, ranging from anatomical to functional assessments. Changes in tumour size are not necessarily correlated with changes in tumour viability and outcome. In addition, morphological changes resulting from therapy occur slower than functional changes. Inclusion of PET images in radiotherapy protocols is desirable because it is predictive of treatment response and provides crucial information to accurately target the oncological lesion and to escalate the radiation dose without increasing normal tissue injury. For this reason, PET may be used for improving the Planning Treatment Volume (PTV). Nevertheless, due to the nature of PET images (low spatial resolution, high noise and weak boundary), metabolic image processing is a critical task. The aim of this Ph.D thesis is to develope smart methodologies applied to the medical imaging field to analyse different kind of problematic related to medical images and data analysis, working closely to radiologist physicians. Various issues in clinical environment have been addressed and a certain amount of improvements has been produced in various fields, such as organs and tissues segmentation and classification to delineate tumors volume using meshing learning techniques to support medical decision. In particular, the following topics have been object of this study: • Technique for Crohn’s Disease Classification using Kernel Support Vector Machine Based; • Automatic Multi-Seed Detection For MR Breast Image Segmentation; • Tissue Classification in PET Oncological Studies; • KSVM-Based System for the Definition, Validation and Identification of the Incisinal Hernia Reccurence Risk Factors; • A smart and operator independent system to delineate tumours in Positron Emission Tomography scans; 3 • Active Contour Algorithm with Discriminant Analysis for Delineating Tumors in Positron Emission Tomography; • K-Nearest Neighbor driving Active Contours to Delineate Biological Tumor Volumes; • Tissue Classification to Support Local Active Delineation of Brain Tumors; • A fully automatic system of Positron Emission Tomography Study segmentation. This work has been developed in collaboration with the medical staff and colleagues at the: • Dipartimento di Biopatologia e Biotecnologie Mediche e Forensi (DIBIMED), University of Palermo • Cannizzaro Hospital of Catania • Istituto di Bioimmagini e Fisiologia Molecolare (IBFM) Centro Nazionale delle Ricerche (CNR) of Cefalù • School of Electrical and Computer Engineering at Georgia Institute of Technology The proposed contributions have produced scientific publications in indexed computer science and medical journals and conferences. They are very useful in terms of PET and MRI image segmentation and may be used daily as a Medical Decision Support Systems to enhance the current methodology performed by healthcare operators in radiotherapy treatments. The future developments of this research concern the integration of data acquired by image analysis with the managing and processing of big data coming from a wide kind of heterogeneous sources.