The Imaging Data Commons (IDC), part of NCI’s Cancer Research Data Commons (CRDC), a cloud-based data science infrastructure that connects cancer research data with analysis tools and resources, gives cancer researchers access to a large collection of medical images and related data. Researchers can use IDC to find CT scans, MRIs, pathology images, expert annotations, and other resources for cancer research. IDC has developed a range of tools for exploring, accessing, downloading, viewing, and analyzing these data. Researchers can also use AI assistants connected to IDC resources to ask questions about the data in natural language.

For example, someone might ask, “Which IDC collections have prostate MRI with expert segmentations?” or “How much CT data is there for lung cancer, and how much of it is cleared for commercial use?” An AI assistant connected to IDC can use IDC’s own tools and information to help answer those questions. When an AI assistant can use tools and take steps to carry out a task, it can also be described as an AI agent.

Dr. Andrey Fedorov, who will demonstrate these capabilities in an upcoming webinar, describes how these agentic interfaces can help researchers work with IDC:

“LLMs already know a lot about DICOM, relevant tools and libraries for image analysis and data wrangling, and SQL for navigating metadata—skills that are essential for using IDC effectively but require time for a cancer researcher to build. With the agentic interfaces to IDC, we complement this general knowledge with the ability to navigate IDC data. The result is amazing - scientists who do not have imaging expertise are finally able to use IDC without extensive training!”

Several related IDC resources support AI-assisted interaction with IDC:

• idc-index is a Python package that provides programmatic access to IDC data and metadata and supports tasks such as searching for and downloading data.
• The IDC agent skill provides an AI assistant with information and instructions for working with IDC and its resources.
• The Model Context Protocol (MCP) interface allows compatible AI assistants to connect directly to IDC tools and data sources. MCP is an open standard for connecting AI assistants with specialized resources.

How researchers use these resources will depend on the task and the AI environment. An AI assistant can work with IDC through the MCP interface or use the IDC skill, which can draw on additional IDC tools and resources. idc-index provides programmatic access to IDC data and metadata. All three build on IDC’s existing infrastructure and can be incorporated into research workflows.  

With an AI assistant connected to IDC resources, some IDC tasks can begin with a straightforward question. A researcher interested in breast cancer, for example, could ask an assistant to find breast MRI data, report the amount of data available, and provide a command for downloading it. A researcher who has assembled a cohort could ask for citations for the collections represented in it.

Researchers still need to understand the data they are working with and determine whether they are appropriate for a particular study. AI assistants can produce convincing but incorrect information, so results should be checked before they are used.

These approaches provide additional ways to interact with IDC’s existing data, metadata, and tools. For researchers who are comfortable starting with a question rather than a database query or command, natural language interaction can be a useful addition to the ways they already explore and work with IDC.

The IDC MCP server and version 3 of the REST API are currently beta releases, so their interfaces may change before the final release. IDC provides information about these tools, setup options, and precautions for using AI assistants in its AI assistants documentation.