Skip to contents

Describe a DICOM Image Using a Vision-Capable LLM

Source: R/llm.R
dicom_describe_image.Rd

Renders the DICOM slice with clinical windowing to a PNG, encodes it as base64, and sends it along with anonymized technical metadata and pixel intensity statistics to a vision-capable LLM (e.g., Ollama llava, OpenAI gpt-4o, or Anthropic claude-sonnet-4-20250514).

Usage

dicom_describe_image(
  path,
  llm = dicom_llm("ollama"),
  frame = 1L,
  window = "auto",
  prompt = NULL
)

Arguments

path

Path to a DICOM file.

llm

A "dicom_llm" object created by dicom_llm().

frame

Integer frame/slice index (1-based). Default is 1L.

window

Window preset ("auto", "brain", "lung", "bone", etc.). Default is "auto".

prompt

Custom instruction prompt for the vision model.

Value

A character string with the image description and quantitative stats.

Examples

dcm_file <- system.file("extdata", "example.dcm", package = "dicomtk")
llm <- dicom_llm("offline")
dicom_describe_image(dcm_file, llm = llm)
#> [1] "DICOM Image Analysis (CT, 512x512, Photometric: MONOCHROME2): Rescaled pixel intensities range from -3024 to 1483 (mean: -704.75, SD: 1230.16, median: -78; 5th-95th percentile: [-3024, 102]). Window Center: 40, Window Width: 400."