AISSLab Presents Groundbreaking 'DeepSpine' AI Research at ICBBBCB-2025 in Turkey
We are proud to announce that members of the Artificial Intelligence and Software Systems Laboratory (AISSLab) presented their cutting-edge research at the prestigious International Conference on Bioinformatics, Biomedicine, Biotechnology and Computational Biology (ICBBBCB-2025). The conference was held in Istanbul, Turkey, from July 3rd to 4th, 2025.
Our researcher, Mukhlis Raza, delivered an insightful presentation on the paper titled, DeepSpine: MCP-Based LLM Agentic Reasoning Chatbot for Automated Lumbar MRI Pathology Analysis and Diagnosis.
The research addresses a significant challenge in modern radiology: the complex and time-intensive analysis of lumbar spine MRIs, which traditionally requires specialized expertise. The team's innovative solution, DeepSpine, is an advanced conversational AI system designed to automate and enhance this process.
The system employs a novel multi-agent framework, where specialized AI agents collaborate on tasks such as image segmentation, measurement extraction, pathology classification, and report generation. This collaboration is orchestrated by a Model Context Protocol (MCP) layer, which ensures all agents maintain a persistent, shared understanding of the diagnostic task. By integrating **Retrieval-Augmented Generation (RAG)**, DeepSpine can also pull in evidence from medical literature to provide well-supported, contextually relevant diagnostic insights.
The results presented at the conference highlight the system's remarkable capabilities:
- High Task Completion: The framework, powered by GPT-4o, achieved a task completion rate of $96.6%$.
- Superior Image Analysis: The ensemble vision models reached a Dice Similarity Coefficient (DSC) of $96.43%$ and an accuracy of $99.66%$ for segmenting key structures in sagittal MRI views.
- Quality Clinical Reports: The system demonstrated sophisticated natural language generation, producing high-quality, clinically relevant reports.
This collaborative effort was led by a talented team from our lab at Sejong University's College of AI Convergence, including Mukhlis Raza, Saied Salem, Afnan Habib, Hyunwook Kwon, and Prof. Mugahed A. Al-antari. The project also highlights a successful international partnership with Dr. Ahmet Arif Aydin from Inonu University, Turkey.
The successful presentation of DeepSpine underscores the significant potential of agentic AI to revolutionize clinical workflows, improve diagnostic accuracy, and support healthcare professionals. AISSLab remains committed to pushing the boundaries of AI in healthcare to solve real-world challenges.
For a detailed look at the methodology and findings, the full paper is available in the conference proceedings: https://academicresearchlibrary.com/proceedings/298
