Skip to content

Medical Imaging Research

Exploring Synthetic Data for Medical AI Research

Investigating image augmentation, segmentation, annotation consistency and model robustness in controlled medical imaging research workflows.

Research methodology only. Not a medical device, no clinical validation, and no Synthex results published yet.

Segmentation studySchematic · not patient data
  • Overlap
  • Predicted only
  • Missed
  • Drawn example, not model output
Schematic segmentation study: a drawn axial slice with a region of interest, its reference mask, and a comparison showing overlap, predicted-only and missed areas. Not patient data and not model output.

Research problem

Why lesion segmentation research is hard

The questions this project investigates, stated as problems rather than solved claims.

Evidence: Conceptual · R-006
  • Limited labelled medical data

    Expert annotation is slow and costly, and access to imaging is tightly governed.

  • Difficult lesion segmentation

    Small, faint or irregular lesions are easy to miss and hard to delineate.

  • Annotation uncertainty

    Experts can disagree on whether, and where, a lesion is present.

  • Domain shift

    Training and deployment data can differ in population, protocol and equipment.

  • Scanner and acquisition variability

    Dose, reconstruction and slice thickness change how images look.

  • Model generalisation concerns

    Good scores on one dataset may not transfer to another.

Research workflow

A research methodology, not a clinical pipeline

The intended experimental workflow. Each stage produces a recorded artefact so results can be reproduced and questioned.

  1. 01

    Research Imaging Dataset

    Public, de-identified collections used under their licence terms (e.g. LIDC-IDRI, CC BY 3.0).

  2. 02

    Preprocessing

    Documented resampling, intensity windowing and region-of-interest extraction.

  3. 03

    Segmentation Masks

    Reference masks derived from expert annotations, with disagreement retained.

  4. 04

    Synthetic AugmentationResearch question

    Controlled variations added to the training split only; never to evaluation data.

  5. 05

    Model Training

    Fixed architecture, configuration and recorded random seeds per run.

  6. 06

    Metric Evaluation

    Dice, IoU, precision, recall and sensitivity on held-out real images.

  7. 07

    Error Analysis

    Failures grouped by lesion size, appearance and acquisition condition.

Evaluation stays real

Synthetic or augmented images are added to training data only. Every reported metric is computed on held-out real images with patient-level splits.

Not a clinical pipeline

This is research methodology. It is not a diagnostic workflow and makes no claim of clinical effectiveness.

Imaging data explorer

What segmentation evaluation looks at

Switch views and evaluation focus to see what a researcher inspects. Original schematic drawings and static text — no patient data, no collection images, no model output.

Illustrative Imaging Explorer — Schematic, Not Patient Data

Illustrative Demonstration
View
Evaluation focus
BOUNDARYSCHEMATIC · NOT PATIENT DATA
  • Overlap
  • Predicted only
  • Missed

Comparison View · Segmentation Boundaries

Overlap (agreement), false-positive area (predicted only) and false-negative area (missed).

What a researcher inspects

  • Where predicted and reference edges diverge
  • Whether errors concentrate on small lesions
  • Under- versus over-segmentation

Metrics used (names only)

  • Dice
  • IoU
  • Boundary distance (e.g. Hausdorff)

Static fixture: an original schematic drawing and descriptive text. No patient data, no images from any collection, and no model output. Not a clinical tool.

Experiment comparison

How configurations would be compared

The comparison protocol for this project. Values appear here only once a result is verified and approved in the evidence register — none are yet.

Comparison protocol

Evidence: Unavailable · R-003

What changes

Nothing — trained on real annotated images only.

Held fixed

Splits, architecture, preprocessing, training budget, seeds.

Question answered

What does the model achieve without synthetic data?

Results matrix

Results by configuration and metric. Cells without verified, approved evidence state that no result is published.
ConfigurationDiceIoUPrecisionRecallSensitivity
Real-data baselineNo verified resultNo verified resultNo verified resultNo verified resultNo verified result
Real + synthetic augmentationNo verified resultNo verified resultNo verified resultNo verified resultNo verified result
Alternative augmentationNo verified resultNo verified resultNo verified resultNo verified resultNo verified result
Challenging evaluation conditionsNo verified resultNo verified resultNo verified resultNo verified resultNo verified result

No results are published: no experiment has been verified and approved in the evidence register. Values will appear here only when they are.

Annotation quality

Auditing the labels before trusting the scores

Evaluation is only as reliable as its reference masks. This is the proposed audit workflow.

  • Missing annotations

    Visible lesions with no mask, which make correct predictions look like false positives.

  • Boundary disagreements

    Masks that differ at the edges, which matters most for small lesions.

  • Suspicious masks

    Empty, fragmented or anatomically implausible shapes.

  • Inter-annotator variation

    Measured agreement between readers, kept rather than averaged away.

  • Expert review queues

    Flagged cases ranked so expert time goes where it matters.

  • Annotation consistency testing

    Re-checks after preprocessing and augmentation to catch drift.

  1. 01

    Automated checks

    Empty, fragmented and implausible masks; agreement between readers.

  2. 02

    Flag and rank

    Cases scored by how likely the label is to be wrong.

  3. 03

    Expert review

    Qualified reviewers accept, correct or reject each flagged label.

  4. 04

    Record decision

    Outcome and rationale stored with the dataset version.

Expert review queue

Illustrative · static fixture
Illustrative review queue with placeholder case labels; contains no patient information
CaseAutomated flagPriority
case-AReaders disagree on lesion presenceHigh
case-BMask fragmented into 3 regionsMedium
case-CBoundary differs between readersMedium
case-DMask empty; model detects a regionHigh

Placeholder case labels only. A real queue would hold no patient-identifiable information on any public surface.

Research limitations

What this research cannot show on its own

  • Synthetic-to-real distribution gaps

    Synthetic images may differ from real scans in ways models exploit.

  • Potential annotation bias

    Reference masks reflect specific readers, protocols and conventions.

  • Dataset representativeness

    Public collections may not represent other populations, scanners or sites.

  • Limited external validation

    Results on one collection need confirmation on independent data.

  • Need for domain expert review

    Clinical experts must review labels, failure cases and interpretations.

  • Technical metrics ≠ clinical effectiveness

    Dice or sensitivity do not establish clinical benefit, safety or readiness.

References

Public research context

Public sources this research direction refers to. Synthex is not affiliated with these organisations, and no images from these collections are reproduced on this site.

  1. [1]
    LIDC-IDRI collection — The Cancer Imaging Archive

    Public thoracic CT collection with radiologist lesion annotations. Licence: CC BY 3.0; TCIA Data Usage Policy applies.

    Evidence: Verified · R-002
  2. [2]
    Armato SG 3rd et al. (2011), Medical Physics 38: 915–931

    The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI) — primary publication.

    Evidence: Verified · R-002
  3. [3]
    Armato III SG et al. (2015), Data From LIDC-IDRI

    Data citation, The Cancer Imaging Archive.

    Evidence: Verified · R-002

Licence: LIDC-IDRI is distributed under CC BY 3.0 subject to the TCIA Data Usage Policy and Restrictions.

Research pilot

Explore Medical Imaging AI Research with Synthex

We welcome research collaborations on annotation quality, robustness evaluation and carefully baselined synthetic augmentation. Please do not send patient data through this website.

No medical device certification, clinical approval, NHS partnership or validated patient outcome is claimed.