Generative AI for Automated Medical Report Generation Using Multimodal Clinical Data

Abstract

The fast rate of Electronic Health Records (EHRs), medical imaging, and clinical documentation have greatly led to the administrative burden on healthcare professionals, which leaves little time to attend to patients directly. Automated creation of medical reports based on Generative Artificial Intelligence (GenAI) has become one of the potential solutions that can enhance the efficiency and consistency of documentation. Nevertheless, the current methods have the shortcomings of only single-modality processing, the absence of domain-specific knowledge, and the possibility of producing clinically inaccurate or hallucinated information. The paper proposes a new multimodal Generative AI architecture that combines structured clinical data, medical images, and physician notes to produce accurate, context-specific medical reports. The suggested system uses a modality-specific encoder, such as a Vision Transformer, to analyze medical images, a ClinicalBERT-based model to interpret text, and an attention-based fusion mechanism to capture a global picture of the patient. A domain-adapted large language model is additionally boosted with clinical knowledge grounded in standardized medical ontologies. To enhance stability, a reinforcement-oriented medical validation layer is proposed that can be used to assess the factual consistency and punish clinically incorrect predictions, which would decrease hallucinations and enhance diagnostic consistency. The proposed approach has proven superior to the currently available baseline methods in linguistic quality, clinical accuracy, and report completeness, while requiring much less time during the documentation phase. Explainability is also used in the system, which connects the generated content with supporting clinical evidence. The suggested framework is a scalable, trustworthy, and privacy-conscious approach to smart clinical documentation that can benefit radiology reporting, discharge summary creation, and real-time clinical decision support. This study is relevant to the creation of reliable and human-friendly Generative AI systems to be used in the next-generation healthcare app.