Skip to main content
Medical imaging teams need annotation accuracy that directly impacts patient outcomes. Avala provides precision annotation tools, multi-stage quality control workflows, and team permission controls that support the rigorous standards clinical-grade training data requires.

Why Avala for Medical Imaging

Medical annotation differs from general computer vision in its tolerance for error — there is essentially none. A missed lesion boundary or misclassified cell type can propagate through the model and affect diagnostic decisions. Avala addresses this with:

Precision Annotation Tools

Polygon and segmentation tools with sub-pixel precision for accurate boundary delineation. Keypoint tools for anatomical landmark placement.

Multi-Stage Quality Control

Configurable review pipelines with spot-checking, targeted review, and full review stages. Support for domain expert reviewers with role-based access.

Team Permissions

Fine-grained access controls to restrict who can view, annotate, and review sensitive medical data. Role-based permissions at the organization, team, and project level.

Audit and Compliance

Task lifecycle tracking from assignment through review and approval. Every annotation action is recorded for audit trail requirements.

Data Types

ModalityAvala Data TypeTypical Annotation
X-rayImageBounding boxes, polygons, classification
CT / MRI slicesImagePolygons, segmentation masks
Pathology slidesImagePolygon regions, classification
Endoscopy videoVideoFrame-level segmentation, tracking
UltrasoundImage, VideoBounding boxes, polygons
Retinal imagingImageSegmentation, classification

Common Tasks

Lesion Detection

Draw bounding boxes or polygons around tumors, nodules, cysts, and other regions of interest. For tasks that require precise boundary delineation (e.g., tumor segmentation for surgical planning), use the polygon tool to trace exact margins. The polygon tool supports:
  • Freeform vertex placement for irregular shapes
  • Edge snapping for clean boundaries
  • Vertex editing to refine placement after initial tracing
  • Sub-pixel accuracy for high-resolution medical images

Organ Segmentation

Create pixel-level segmentation masks for organs and anatomical structures in CT or MRI slices. Use the segmentation brush for large regions and switch to polygon mode for fine boundary work.
For organ segmentation tasks, define your label taxonomy with clear hierarchy: organ system > organ > substructure. For example: cardiovascular > heart > left_ventricle. This makes the annotation process faster and the resulting data more useful for model training.

Cell Classification

Classify cell types in pathology slides using classification labels and structured attributes. Define a taxonomy that includes:
  • Primary cell type (e.g., lymphocyte, neutrophil, epithelial)
  • Morphological attributes (e.g., size, shape regularity, staining intensity)
  • Diagnostic relevance (e.g., normal, atypical, malignant)
Classification can be applied at the object level (individual cells) or scene level (tissue regions).

Surgical Video Analysis

Track surgical instruments and anatomical landmarks across endoscopy or surgical video frames. Object tracking maintains consistent IDs across frames, making it possible to train models for instrument detection, phase recognition, and activity analysis.

Quality Control for Medical Data

Medical annotation quality control goes beyond general-purpose review. Avala’s quality control features support the workflows medical teams require.

Multi-Stage Review Pipelines

Configure review pipelines that match your clinical validation process:
Annotation (technician)
  -> First review (trained annotator)
  -> Expert review (radiologist / pathologist)
  -> Approved
At each stage, reviewers can approve, reject with comments, or flag specific annotations with issues. Rejected tasks return to the annotator with clear feedback.

Annotation Issues

Pin issues to specific annotations in the image. A reviewer can mark a polygon boundary as “too loose at the superior margin” and the annotator sees the issue pinned to the exact location that needs correction.

Consensus Workflows

For validation datasets and ground truth creation, assign the same images to multiple domain experts independently. Consensus scoring reveals:
  • Regions where experts disagree (these need additional review or clearer guidelines)
  • Annotators who consistently deviate from the group
  • Edge cases where the annotation guideline is ambiguous

Quality Metrics

Monitor annotation quality across your team:
MetricWhat It Measures
Acceptance ratePercentage of tasks approved on first submission
Annotation timeAverage time per task — unusually fast or slow may indicate issues
Issue frequencyRate of flagged problems per task
Inter-annotator agreementConsistency across annotators on the same data

Compliance Considerations

Medical imaging data often falls under regulatory requirements (HIPAA, GDPR, MDR). While Avala provides the tooling for annotation workflows, your team is responsible for ensuring data handling complies with applicable regulations. Avala features that support compliance workflows:
RequirementAvala Feature
Access controlRole-based team permissions restrict data access to authorized users
Audit trailTask lifecycle tracking records every annotation, review, and status change
Data isolationDatasets and projects are scoped to organizations with membership controls
Export controlExports are generated on demand and can be restricted by permission
Avala does not provide HIPAA BAA or DICOM integration out of the box. If your workflow requires these, contact support@avala.ai to discuss your compliance requirements before uploading protected health information.

Avala Features Used

FeaturePurposeLearn More
Polygon annotationPrecise boundary delineation for lesions and organsPolygon Tool
Segmentation annotationPixel-level masks for anatomical structuresSegmentation Tool
Keypoint annotationAnatomical landmarks for pose and structureKeypoint Tool
ClassificationCell type and tissue classificationClassification Tool
Multi-stage reviewConfigurable review pipelines with expert reviewersQuality Control
Team permissionsRestrict access to sensitive dataTeam Permissions
API exportsIntegration with training pipelinesExports API
Work batchesDistribute annotation work across teamsWork Batches

Getting Started

1

Set up your organization

Create an organization and invite your annotation and review teams. Configure team roles so that only authorized users can access medical data.
2

Upload imaging data

Create a dataset with the appropriate data type (Image for radiology/pathology, Video for endoscopy) and upload your files.
3

Define your label taxonomy

Set up object classes, attributes, and classification categories that match your clinical annotation guideline. Include clear definitions and reference examples.
4

Configure quality control

Set up a multi-stage review pipeline. Assign domain expert reviewers and configure acceptance criteria.
5

Annotate, review, and export

Annotators label the data, reviewers validate at each stage, and you export the approved annotations for model training.

Next Steps