Transformative Medical Report Generation for Brain MRI using BERT-GPT4 Hybrid Models and Real-Time Classification

Abstract

The need for automated medical report generation is critical in healthcare, especially in the radiology domain where generating structured, coherent, and clinically accurate brain MRI reports remains labor-intensive in process. Most of the available methods cannot contextually interpret sparse keyword input or accurately categorize report segments, which limits their utility for real-world clinical applications. In light of these limitations, we propose an advanced multi-module framework that integrates a transformer-based generative model, real-time text categorization, and quality assurance. The hybrid BERT-GPT4 architecture combines the robust clinical semantic-encoding abilities of BERT with GPT4’s fluency in text generation capabilities to generate high-quality medical reports from keyword inputs. This architecture allows for accurate expansion from medical terms such as “ischemic stroke” or “white matter lesion” to coherent report narratives, aiming at more than 85% coherence and above 90% clinical relevance. To implement real-time report categorization, we employ an enhanced hierarchical classification model using a DistilBERT, in which an attention mechanism can serve for efficient and accurate classification in the sections History, Physical Examination, and Findings with more than a 92% F1-score. Moreover, we utilize MedSpacy for keyword extraction and standardization, with lexical analysis and clinical validation to meet medical standards such as SNOMED CT and ICD-10. This ensures syntactic coherence and clinical relevance, achieving scores over 95% and 98%, respectively. This well-rounded approach not only accelerates report generation but also improves accuracy and readability, promising a significant impact on the efficiency of clinical documentation and diagnostic support. Our model processes with a processing time of less than three seconds. Our model will suit applications that have a need to function in real-time and also offers an efficient, reliable tool in the modern medical reporting process.