Use of Generative AI to Support Structured Radiological Reporting in Cardiac Magnetic Resonance Imaging

Free-text reporting is still widely used in radiology, but it often suffers from a lack of completeness, standardization, and comparability, making automation and integration with hospital systems difficult. Structured reporting, on the other hand, provides a standardized framework that improves clarity, consistency, and clinical communication. This thesis presents the design and development of 2GEN4MED, a full-stack web application that allows freelance radiologists to manage clinical appointments and generate structured reports for cardiovascular magnetic resonance (CMR) studies. The system integrates a locally deployed generative AI model to automatically produce reports by combining multiple data sources: user-provided inputs, information extracted from DICOM files, and parameters derived from published reference studies. The resulting reports are standardized according to SCMR guidelines, ensuring that essential clinical information is included in a clear and consistent format. This approach reduces variability, minimizes manual work, and improves workflow efficiency for radiologists. The project demonstrates that the integration of structured reporting and generative AI within a web-based platform can support radiologists in their daily practice, enhancing both the quality of reporting and the usability of clinical data. Future developments could include integration with hospital information systems, role-based access, multilingual support, and the adoption of larger AI models to further improve accuracy and interoperability