Medical image classification training data

It assigns expert-defined categories to medical images or volumes so a model can learn whole-item labels such as anatomy, finding presence, or image quality.

Classification labels the image or volume as a whole. Segmentation marks the boundaries of a region within the image. A dataset can include both when required.

Yes. Define separate properties and multi-select choices where your protocol allows more than one category to apply.

Yes. Include an uncertainty category or another task-defined property and document how reviewers should use it consistently.

Yes. Define quality categories based on visible artifacts or your dataset’s suitability criteria, then have specialists apply and review those labels.

Yes. Supported medical region annotations can accompany item properties when your protocol requires both the category and its visible location.

Unitlab supports DICOM, NIfTI, and NRRD inputs. DICOM slices are grouped by series into an annotation volume, with available views determined by the imaging data.

Use a review stage to inspect labels against the source images, record comments, and return corrections through the workflow. Your team defines the annotation protocol and expert review criteria.

Yes. Versioned releases and supported exports let teams deliver reviewed annotations to downstream pipelines. Select an export that preserves the geometry and metadata your research task requires.