Lung cancer is one of the most common cancers and has one of the worst prognoses, mainly due to the difficulty of early diagnosis. In Italy, there are an estimated 41,000 new cases each year, and in 2021, the disease was responsible for approximately 34,000 deaths. The social impact is significant, as the disease is often diagnosed at an advanced stage, when the chances of survival are reduced: the 5-year survival rate is around 18% in advanced stages, while it can reach 90% if diagnosed at an early stage.
Early-stage lung cancer mainly manifests itself in the form of pulmonary nodules, which can be detected by computed tomography (CT). However, the diagnosis of these nodules often requires invasive procedures, such as bronchoscopy, CT-guided needle biopsy, or surgical biopsies, which affect patients' quality of life and healthcare costs. For this reason, the ability to accurately distinguish between benign and malignant nodules is a central theme in clinical research.
In recent years, artificial intelligence, particularly deep learning techniques, has shown considerable potential in supporting CT screening. Results show that AI can achieve performance superior to that of individual radiologists and comparable to that of a multidisciplinary team, using histological reports as a diagnostic reference. This confirms the value of AI as a tool to support clinical decision-making.
Considering the multimodal nature of clinical data (images, text reports, diagnostic tests), there is growing interest in models capable of integrating multiple sources of information. In this context, the research project aims to develop a system capable of automatically recognizing pulmonary nodules and generating natural language text descriptions of the findings.