Bits & Flames
bitsandflames/fyron

About Fyron

Fyron is built by Bits & Flames for clinical data science teams who need transparent, interoperable, paper-ready analysis workflows.

Our slogan is simple: Built by bits. Shaped by flames. The bits are the careful engineering details: schemas, tables, masks, metrics, file hashes, and reproducible runs. The flames are the spark of scientific work: asking better questions, testing assumptions, and turning messy healthcare data into evidence that can be reviewed.

Who We Are

Bits & Flames builds tools for healthcare data, imaging, AI, and scientific workflows. Fyron is our open toolkit for the practical layer between clinical systems and clinical evidence: extracting data, shaping cohorts, validating models, and producing outputs that are understandable to engineers, clinicians, and reviewers.

We care about software that feels useful in the real world:

  • explicit instead of opaque,
  • modular instead of monolithic,
  • reproducible instead of notebook-only,
  • scientific instead of decorative,
  • easy to inspect before it is trusted.

What We Want Fyron To Do

Fyron should make high-quality clinical data science less fragile. A typical project may need FHIR resources, DICOM images, segmentation masks, PDF reports, cohort joins, survival endpoints, ML models, validation figures, and publication tables. Fyron keeps those steps connected while preserving clear boundaries.

The goal is not AutoML and not automatic scientific interpretation. The goal is a toolkit that helps researchers and data scientists build defensible pipelines:

  1. Collect data from clinical systems.
  2. Build explicit patient-level cohorts.
  3. Extract imaging, BOA, radiomics, and document-derived features.
  4. Run descriptive analysis, statistics, survival analysis, and ML.
  5. Validate thresholds, calibration, subgroups, survival scores, and segmentation quality.
  6. Export paper-ready plots, tables, and provenance manifests.

What Makes Fyron Different

Fyron is designed around inspectable outputs:

PrincipleHow Fyron Applies It
DataFrame-firstClinical tables stay easy to inspect, join, save, and review.
Explicit assumptionsColumn names, thresholds, event definitions, labels, and paths are passed directly.
Optional heavy dependenciesImaging, survival ML, radiomics, and DICOM SEG tools stay behind extras.
Matplotlib-native plottingFigures remain saveable, editable, and manuscript-friendly.
ReproducibilityCLI commands and audit manifests make reruns easier to document.
Clinical cautionOutputs support analysis; they do not replace clinical validation.

Who Fyron Is For

Fyron is useful for:

  • clinical data scientists building cohort studies,
  • imaging AI teams working with DICOM, NIfTI, DICOM SEG, BOA, or radiomics,
  • researchers preparing descriptive analyses, survival models, and validation tables,
  • engineers building repeatable ETL and analysis jobs,
  • LLM/agent workflows that need clear module boundaries and explicit tools.

The Fyron Shape

Fyron works best when each step is small and reviewable:

python
features = extract_features(...)
cohort = join_cohort_tables(features, outcomes, on="patient_id")
cohort = build_survival_columns(cohort, "index_date", "last_followup", "event")
summary = descriptive.create_table_one(cohort, group_col="risk_group")
validation = validation.calibration_summary(y_true, y_prob)
manifest = audit.create_provenance_manifest(inputs=[...], outputs=[...])

That shape is deliberate. It keeps clinical assumptions visible and leaves room for domain expertise at every decision point.

Where To Go Next