Digital Transformation of Medical Imaging : from quantitative diagnostic to surgical training

Abstract

Medical imaging is the prominent way of knowing internal structure, morphology and functional aspects of a living human body and its widely adopted in clinical practice to diagnose, support and plan treatments. This thesis explores the digital transformation of medical imaging, demonstrating how pixel-intensity analysis and advanced computational methodologies can enhance diagnostic precision and support surgical training. Leveraging data-rich imaging modalities such as computed tomography (CT) and infrared imaging, this work develops and validates quantitative pipelines to extract robust imaging biomarkers and reconstruct anatomically accurate models for both clinical and educational purposes. From a diagnostic perspective, novel radiodensitometric and radiomic techniques were applied across three clinical domains. First, CT-based quantitative analysis was used to characterize skeletal muscle aging, introducing biomarkers of sarcopenia and fat infiltration. Second, the concept of “virtual cardiac histology” was developed, where high-dimensional radiomic features from cardiac CT enabled the non-invasive differentiation of healthy myocardium from pathological conditions, including hypertrophic cardiomyopathy (HCM) and acute myocardial infarction (AMI). Third, automated meibography analysis was proposed to quantify morphological changes in Meibomian glands from infrared eyelid imaging, offering objective metrics for ocular surface diseases. From an educational and surgical standpoint, this research introduces the Radio Anatomical Interactive Library (RAIL), an innovative platform that integrates clinical imaging data, 3D reconstructions, and mixed reality tools to support anatomical learning, preoperative planning, and surgical rehearsal. By combining radiological data with immersive technologies, RAIL enhances comprehension of complex anatomies and promotes interactive, case-based training. Collectively, the studies presented demonstrate that pixel-level analysis of medical images can yield reproducible and clinically meaningful biomarkers, bridging the gap between qualitative interpretation and quantitative, data-driven diagnostics. This thesis highlights the potential of integrating AI-based radiomics, 3D visualization, and extended reality into the healthcare continuum, enabling personalized diagnostics, improving training efficiency, and advancing the paradigm of precision medicine.