Development of an Agentic AI Framework for Multi-Specialty Medical Imaging Diagnostics: Leveraging MedGemma for Enhanced Clinical Decision Support
Abstract: The integration of artificial intelligence (AI) in medical imaging has revolutionized diagnostic workflows, yet challenges such as data imbalances and misdiagnosis risks persist. This paper presents an innovative agentic AI framework utilizing Google's MedGemma-27b-it model for multi-specialty medical imaging diagnostics. Drawing from MedMNIST datasets across pathology, chest X-ray, oncology, and pneumonia imaging, the framework simulates diagnostic processes, identifies imbalances, and generates actionable recommendations through iterative agent loops. Key innovations include a TPU-sharded model deployment for scalability and a simulation of post-intervention accuracy improvements from 91.39% to 94.15%. Mathematical formulations for prevalence and misdiagnosis risk are introduced, alongside algorithms for data augmentation and bias mitigation. Experimental results demonstrate the framework's efficacy in handling 277,656 images, with powerful use cases in addressing radiologist shortages and enabling predictive analytics in U.S. healthcare. This work underscores the importance of agentic AI in enhancing diagnostic reliability, potentially reducing misdiagnoses by up to 20% in high-risk classes like pneumonia X-rays.
Keywords: agentic AI, medical imaging, MedGemma, data imbalance, diagnostic accuracy, healthcare AI
How to Cite?: Karan Chandra Dey, "Development of an Agentic AI Framework for Multi-Specialty Medical Imaging Diagnostics: Leveraging MedGemma for Enhanced Clinical Decision Support", Volume 14 Issue 8, August 2025, International Journal of Science and Research (IJSR), Pages: 724-727, https://www.ijsr.net/getabstract.php?paperid=SR25812175614, DOI: https://dx.doi.org/10.21275/SR25812175614