AI-Assisted Organoid Image Analysis: Evaluating a Model Context Protocol (MCP) Server Against Traditional Pipeline in Traumatic Brain Injury (TBI) Research*
Session Number
2
Advisor(s)
John Finan (PI), Shahrzad Shiravi (PhD Candidate), and Yasaman Samei (PhD student), University of Illinois at Chicago
Location
B108
Discipline
Engineering
Start Date
15-4-2026 11:10 AM
End Date
15-3-2026 11:55 AM
Abstract
Brain organoids made from human induced pluripotent stem cells (iPSCs) became a valuable model for studying traumatic brain injury (TBI). To understand effects at the cellular level, researchers rely on accurate image and video analysis. Traditional tools such as Napari use hardcoded Python to analyze variables (e.g., size and orientation) of objects in images. However, it requires a developed skill set in programming, which limits access for researchers without coding experience. Recent advances in artificial intelligence (AI), including large language models such as Claude, offer new ways to interact with research software. Through the Model Context Protocol (MCP), AI systems connect directly with tools using natural language instructions. This study evaluates whether an MCP server-based pipeline can perform comparably to hardcoded Napari scripting for brain organoid image analysis. This study focused on the technique “annotated segmentation,” available in Napari as a Python file and in MCP as natural language prompting using Claude via the server. Both pipelines used a subset of organoid images, and performance was compared through segmentation accuracy, efficiency, and consistency. This study aims to assess the viability of MCP-based tools as an opening into biomedical image analysis and their potential role in advancing research in TBI.
AI-Assisted Organoid Image Analysis: Evaluating a Model Context Protocol (MCP) Server Against Traditional Pipeline in Traumatic Brain Injury (TBI) Research*
B108
Brain organoids made from human induced pluripotent stem cells (iPSCs) became a valuable model for studying traumatic brain injury (TBI). To understand effects at the cellular level, researchers rely on accurate image and video analysis. Traditional tools such as Napari use hardcoded Python to analyze variables (e.g., size and orientation) of objects in images. However, it requires a developed skill set in programming, which limits access for researchers without coding experience. Recent advances in artificial intelligence (AI), including large language models such as Claude, offer new ways to interact with research software. Through the Model Context Protocol (MCP), AI systems connect directly with tools using natural language instructions. This study evaluates whether an MCP server-based pipeline can perform comparably to hardcoded Napari scripting for brain organoid image analysis. This study focused on the technique “annotated segmentation,” available in Napari as a Python file and in MCP as natural language prompting using Claude via the server. Both pipelines used a subset of organoid images, and performance was compared through segmentation accuracy, efficiency, and consistency. This study aims to assess the viability of MCP-based tools as an opening into biomedical image analysis and their potential role in advancing research in TBI.