The Task
The client needed labeled images to train and validate a computer vision model that classifies female pattern hair loss. Each participant provided two photos (a top view and a frontal view) showing the scalp, hairline, and forehead area.
The dataset was annotated using the Sinclair Scale, which includes five stages of female alopecia. The main target was to collect balanced data for Stages 2, 3, and 4, with 50 validated samples per stage.
Key challenges included:
- Lower participation rates compared to standard biometric tasks
- Difficulty separating borderline cases between Sinclair stages
- Budget limits
- Privacy concerns when collecting identifiable facial data
The Solution
Preparation and guidelines
- Selected the Sinclair Scale as a clinically relevant and easy-to-understand framework for female alopecia assessment
- Built visual-first guidelines using real examples for each stage rather than abstract descriptions
- Added explanations for common participant mistakes, especially confusion between adjacent stages
- Included borderline and transitional cases to clarify where one stage ends and another begins
Data collection process
- Introduced eye-area blurring to reduce biometric sensitivity
- Optimized task pricing to balance budget constraints and sustainable collection speed
- Focused collection on Sinclair Stages 2, 3, and 4, while continuing to review all incoming submissions
- Identified and collected high-quality samples from Stages 1 and 5
Annotation and quality control
- Reviewed all submissions through an experienced assessor with prior alopecia annotation background
- Carefully filtered and reclassified ambiguous cases based on expert judgment
- Delivered only high-confidence, clearly validated samples in the final dataset
| Stage | Input | Workflow Scope | Main Quality Checks |
|---|---|---|---|
| Participant Screening | Crowd platforms | Targeted recruitment, eligibility filtering | Stage relevanceParticipation rate |
| Image Collection | Raw scalp images | Capture of top and frontal views | Scalp visibilityImage clarity |
| Privacy Processing | Collected images | Eye-area blurring, PII risk reduction | Face anonymizationVisual usability |
| Pre-Validation | Image pairs | Manual and automatic checks | View completenessTechnical quality |
| Expert Annotation | Validated images | Sinclair Scale labeling | Stage separationBorderline handling |
Audience Screening
5 days
Pilot Collection & Validation
1 week
Targeted Collection & Expert Annotation
6 weeks
Final QA & Dataset Delivery
1 week
The Results
- Collected and annotated 150+ female alopecia cases
- Met the required distribution across Sinclair Stages 2, 3, and 4
- Expanded the dataset with validated Stage 1 and Stage 5 examples
- Delivered a clean, well-structured dataset ready for medical AI training
The key challenge in female alopecia projects is sourcing relevant participants at scale. Quality depends on focused recruitment, strict entry checks, and expert review of every sample.
- Hanna Parkhots
- Data Collection Team Lead
