Unveiling healthcare data archiving: Exploring the role of artificial intelligence in medical image analysis

Medical digital archives can be seen as contemporary databases designed to store and manage vast amounts of medical information, from patient records and clinical studies to medical images and genomics data. The structured and unstructured data that compose the archives undergo rigorous curation processes, to ensure their accuracy, reliability, and standardization for clinical and research purposes. In the rapidly evolving field of healthcare, artificial intelligence (AI) is emerging as a transformative force, able to reform medical digital archives improving the management, analysis, and retrieval of vast clinical datasets, and ultimately leading to more informed decisions, timely interventions, and improved patient outcomes. Specifically, managing medical images in digital archives poses numerous challenges such as data heterogeneity, image quality variability and lack of annotations, that can be addressed with AI solutions. This thesis aims to exploit AI algorithms for the analysis of medical images stored in digital archives. This work investigates various medical imaging techniques, each of which is characterized by a specific application domain and consequently presents a unique set of challenges, requirements, and potential outcomes. In particular, it delves into AI diagnostic assistance for three critical imaging techniques in specific clinical scenarios: i) Endoscopic imaging obtained during laryngoscopy examinations; this includes in-depth exploration of techniques such as keypoint detection for vocal fold motility estimation and upper aerodigestive tract cancer segmentation; ii) Magnetic resonance imaging for intervertebral disc segmentation, for the diagnosis and treatment of spinal conditions and diseases, as well as image-guided interventions; iii) Ultrasound imaging in rheumatology, for carpal tunnel syndrome evaluation through median nerve segmentation. The methodologies presented in this work demonstrate the feasibility of using AI algorithms for the analysis of archived medical images, and the achieved methodological advances highlight the potential of AI-based algorithms in extracting useful information implicitly contained in digital archives.