Configure an LLM Provider for DICOM Analysis

Creates a "dicom_llm" configuration object connecting dicomtk to:

Local Ollama (provider = "ollama", default http://localhost:11434)
OpenAI / OpenAI-compatible APIs (provider = "openai" or "custom", supporting OpenAI, OpenRouter, LM Studio, vLLM, Azure, or local proxy subscriptions)
Anthropic Claude (provider = "anthropic", supporting native /v1/messages as well as OpenAI-compatible endpoints)
Offline rule-based engine (provider = "offline", for deterministic local metadata & pixel-statistics summaries without external API calls)

Usage

dicom_llm(
  provider = c("ollama", "openai", "anthropic", "custom", "offline"),
  model = NULL,
  base_url = NULL,
  api_key = NULL,
  temperature = 0.2,
  max_tokens = 1024L,
  anonymize = TRUE,
  offline_fallback = TRUE
)

Arguments

provider: Character provider type: "ollama", "openai", "anthropic", "custom", or "offline". Default is "ollama".
model: Model identifier (e.g., "llava:13b", "llama3.2-vision", "gpt-4o", "claude-sonnet-4-20250514").
base_url: Optional API base URL. Defaults to "http://localhost:11434" for Ollama, "https://api.openai.com/v1" for OpenAI, and "https://api.anthropic.com/v1" for Anthropic.
api_key: API key or bearer token. Automatically reads from OPENAI_API_KEY or ANTHROPIC_API_KEY environment variables if not supplied.
temperature: Sampling temperature. Default is 0.2.
max_tokens: Maximum response tokens. Default is 1024L.
anonymize: Logical. Automatically strip HIPAA PHI tags before sending metadata to non-local endpoints. Default is TRUE.
offline_fallback: Logical. If TRUE (default), falls back to a deterministic local clinical summary if the remote/local server is unreachable.

Value

An S3 object of class "dicom_llm".

Examples

# Local Ollama configuration
local_llm <- dicom_llm("ollama", model = "llava:13b")

# OpenAI-compatible configuration
openai_llm <- dicom_llm("openai", model = "gpt-4o")

# Anthropic Claude configuration
claude_llm <- dicom_llm("anthropic", model = "claude-sonnet-4-20250514")