CaPTion 2026 Keynote
Kostas Marias, PhD
Professor of Medical Image Processing, Hellenic Mediterranean University
Biography
Kostas Marias, PhD, is Professor of Medical Image Processing in the Department of Electrical and Computer Engineering at the Hellenic Mediterranean University (HMU), Dean of the HMU School of Engineering, and Head of the Computational BioMedicine Laboratory at the Institute of Computer Science, Foundation for Research and Technology–Hellas (FORTH-ICS). He received his PhD in Medical Image Analysis and Medical Physics from UCL’s Royal Free and University College Medical School, in collaboration with the University of Oxford, where he subsequently worked as a Research Assistant. In 2026, he was appointed Visiting Professor and Distinguished International Scholar “Chiara Fama” at the University of Torino.
Professor Marias has coordinated major European projects in cancer modelling and personalised medicine, including ContraCancrum and TUMOR, and has led or contributed to numerous international initiatives in medical imaging, biomedical informatics, and AI for healthcare. A Senior Member of IEEE, he has authored or co-authored more than 350 publications spanning medical image processing, image-based modelling, radiomics, and deep learning for cancer. He serves as Section Editor-in-Chief for Medical Imaging at the Journal of Imaging and is co-founder of SYNTHAINA.AI, a company specialising in synthetic data generation for AI.
Keynote Talk
Linking Medical Imaging to Cancer Biology Through AI: From Detection to Clinical Translation
Talk Abstract
Artificial intelligence in medical imaging has achieved remarkable performance in detecting, segmenting, and assessing tumors. Yet cancer is not a single imaging target, but a heterogeneous and evolving biological system. Because precision oncology increasingly relies on molecular profiling to classify tumors and guide treatment, medical imaging AI must progress beyond tumor detection toward revealing clinically relevant aspects of cancer biology. Rather than replacing pathology or genomic testing, AI-enabled imaging can complement them by providing a noninvasive, whole-tumor, and longitudinal view of cancer from detection and biological characterization to treatment selection and clinical translation.
This talk will explore how quantitative imaging, radiomics, deep learning, and multimodal integration can link imaging phenotypes with tumor biology. Examples from our research include multiparametric breast MRI for characterizing tumor heterogeneity, vascular organization, molecular phenotypes, and treatment response; lung CT for the noninvasive prediction of molecular characteristics, including EGFR and KRAS mutation status; and multiparametric brain MRI for predicting IDH1 mutation status in gliomas. Clinical translation, however, requires more than promising predictive performance. Imaging–biology associations must remain reliable across scanners, acquisition protocols, institutions, and patient populations. Achieving this requires biologically motivated, robust, interpretable, and externally validated AI.
