Advancing Vertigo diagnosis with large language models: A multimodal, AI-driven approach to Etiology differentiation

Abstract

Vertigo is a common and often debilitating condition with various underlying causes, including vestibular disorders, neurological conditions, and systemic diseases. Accurate diagnosis and differentiation of vertigo etiology remain challenging due to the diverse and overlapping symptoms associated with various situations. This research presents an AI-driven approach for the diagnosis and etiology prediction of vertigo, leveraging Large Language Models (LLMs) such as GEMMA and LLaMA. By integrating multimodal data, including patient history, symptoms, results from otoneurologic tests, audiologic assessments, and imaging findings, this research aims to enhance the precision and efficiency of vertigo diagnosis. The data set used in this study includes comprehensive information on patient medical histories, clinical test results, and imaging outcomes, which are processed and analyzed by AI-driven models to classify and predict vertigo causes. We evaluate the performance of different machine learning and deep learning models, including LLMs, to identify the most effective approach for vertigo diagnosis. The results demonstrate the promising capabilities of AI in improving diagnostic accuracy, offering a powerful tool for clinicians to differentiate between various types of vertigo and optimize treatment strategies. This research contributes to the growing literature on AI-assisted diagnosis in otoneurology, paving the way for more accurate, personalized, and timely vertigo management. The GEMMA model achieved an accuracy of 92%, while the LLaMA model surpassed this with an accuracy of 94%. An ensemble method combining both models produced the highest accuracy at 96%, demonstrating the advantages of model fusion in improving diagnostic performance. The results indicate that using LLaMA and GEMMA, particularly in combination, provides a powerful tool for precise vertigo differentiation, helping clinicians make informed decisions for better patient outcomes. This approach highlights the potential of LLMs in advancing medical diagnostics, offering promising implications for developing AI-driven diagnostic systems in healthcare.