Your first investigation

Use Python 3.11–3.14 and a pandas dataframe. Fieldwork needs pandas and NumPy; it does not require DICOM tools, a plotting server or a notebook.

Install Fieldwork 0.1.0

pip install fieldwork==0.1.0

From a uv project, use uv add fieldwork==0.1.0. This is the initial alpha release; see Documentation source for its exact tag and commit.

Make a small dataset

import pandas as pd
import fieldwork as fw

df = pd.DataFrame({
    "site": ["North", "North", "South", "South"],
    "exam": [1, 1, 2, 2],
    "image_1": [10, 11, 20, 21],
    "image_2": [12, 13, 22, None],
}, index=[0, 0, 0, 0])

Each exam has two rows. The second image is missing on the last row. Duplicate indexes are intentional.

Orient, then choose a path

overview = fw.explore(df)
print(overview)
paths = fw.suggest_paths(df, features=["site", "exam"])
if paths.best is not None:
    tree = paths.best.census(df)
    print(tree)

Site and exam are equivalent groupings in this small example; both describe two groups. In a larger dataset with several exams per site, supported nesting favors site before exam. Explicit dimensions always work: fw.census(df, ["site", "exam"]).

Inspect the missing image

availability = fw.missingness(df, min_implication=0.75)
edge = next(
    f for f in availability["findings"]
    if f["pattern"] == "presence_implication"
    and [ref["column"] for ref in f["features"]] == ["image_1", "image_2"]
)
exceptions = availability.inspect(df, edge["id"], exceptions=True)
assert exceptions["exam"].tolist() == [2]

The measured conditional presence is 3/4. Continue with families and exceptions, or open the worked notebook.