/MediVQA

Part of our final year project work involving complex NLP tasks along with experimentation on various datasets and different LLMs

Primary LanguageHTML

In this work, a LoRA-optimized BLIP model is used to investigate the improvement of Medical Visual Question Answering (VQA). Our work aims to optimize the BLIP architecture through the use of Low-Rank Adaptation (LoRA) to develop a more effective and resource-efficient method for medical image analysis. We rigorously test this technique on a specialized combination of medical VQA datasets and show its efficacy. The outcomes demonstrate notable gains in accuracy, especially for closed-type questions, highlighting the potential of LoRA-enhanced BLIP models to advance AI-driven healthcare solutions and medical diagnostics. This paper lays the groundwork for future research and development in the field by presenting a novel approach that connects cutting-edge AI techniques with essential medical applications.