Recent advances in artificial intelligence suggest that performance gains in medical imaging arise not only from improved classifiers but from learning more meaningful internal representations of disease. Building upon our work on thyroid nodule malignancy prediction using contrastive self-supervised learning and curriculum-based fine-tuning, this seminar explores a broader perspective on the future of medical AI. We discuss how contrastive learning restructures the latent feature space of ultrasound images, leading to clinically meaningful organization of benign and malignant nodules. Extending this idea, we examine emerging concepts from Large Concept Models (LCMs), which propose reasoning at the level of semantic concepts rather than individual tokens, and from multi-agent AI systems, where specialized agents collaborate to solve complex tasks.
By connecting representation learning, concept-centric reasoning, and agent-based intelligence, we propose a future framework for thyroid ultrasound analysis in which AI systems move beyond simple classification toward interpretable clinical concept discovery and collaborative diagnostic reasoning. Such systems may provide improved transparency, robustness, and clinical trust while opening new directions for AI-assisted medical research and decision support. The presentation aims to bridge current advances in self-supervised learning with emerging paradigms in AI research and to discuss their potential impact on the next generation of explainable medical intelligence.
Short Bio:
I am an Assistant Professor of Endocrinology at the Victor Babeș University of Medicine and Pharmacy Timișoara, specialist in Endocrinology and Diabetes, Nutrition and Metabolic Diseases, and holder of a PhD in Medicine focused on the molecular and angiogenic mechanisms of pituitary tumors. My clinical and research activity spans endocrine oncology, thyroid disease, adrenal disorders, metabolic diseases, and medical ultrasonography.
Driven by a growing interest in the transformative potential of artificial intelligence in healthcare, I pursued a Master’s degree in Bioinformatics at the West University of Timișoara, where I became increasingly involved in machine learning, representation learning, and computational approaches to biomedical data. This interdisciplinary journey led me to begin a second PhD, in Artificial Intelligence, at the West University of Timișoara.
My current research focuses on explainable and multimodal AI systems that integrate imaging, clinical, metabolic, and molecular data to support decision-making in complex endocrine diseases. I am particularly interested in developing interpretable machine learning models capable of identifying clinically meaningful phenotypes and advancing personalized medicine.
I envision a future in which medicine and computer science converge into a unified discipline, where artificial intelligence serves not only as a predictive tool but as a collaborative partner in diagnosis, prognosis, and scientific discovery, enabling more precise, personalized, and patient- centered healthcare.
