How MyZenCheck's 7 Tongue-Image Components Work
A code-aligned overview of the image-quality check, six feature-analysis components, Azure upload, synthesis, and important interpretation limits.
TL;DR
The current code performs one image-quality check and six visual feature analyses before software generates TCM-oriented educational text. The output is not a medical diagnosis, and component confidence is not clinical accuracy.
Quick answer
MyZenCheck uses seven computer-vision components in the current flow: one initial image-quality/tongue-presence check and six components that return labels for visible characteristics. Additional software combines those outputs into educational TCM-oriented text.
This architecture does not make the product a medical device, and it does not prove diagnostic accuracy.
1. Browser capture and quality check
The browser sends image bytes over HTTPS to the quality-check endpoint. The first model evaluates whether a tongue is visible and whether the image appears usable. Poor focus, lighting, framing, or tongue visibility may trigger a new-photo request.
2. Azure upload
For an accepted image, the browser requests a time-limited write URL and uploads the file to Azure Blob Storage. The application then sends the resulting blob URL to the analysis endpoint.
The photo therefore does not stay only on the device. After an analysis request finishes, the API attempts to delete the submitted blob and logs cleanup failures. Immediate deletion cannot be guaranteed, and abandoned uploads still require a platform lifecycle backstop. See the privacy policy.
3. Six visual feature components
The API calls six configured Custom Vision endpoints. Their internal code labels refer to shape, location, edge/surface, coating, color, and moisture-like characteristics. These are image-label tasks; they do not measure organ function or determine the cause of a feature.
4. Educational synthesis
An Azure-hosted text-generation service receives the model tags and produces an explanation using TCM pattern vocabulary. Generated text can be wrong, overconfident, or incomplete. It must stay within the product’s educational and safety boundaries.
5. Browser result
The result is returned to the browser and used by the result interface. Browser storage may retain result history for the user flow. A model confidence value should not be interpreted as the probability that a health conclusion is correct.
What is not established
MyZenCheck has not published sufficient evidence to establish diagnostic accuracy, clinical utility, treatment effectiveness, representative subgroup performance, or independent clinical validation. Read the evidence and validation status and accuracy limitations guide.
Key Takeaways
- ✓ A quality check runs before upload and detailed analysis
- ✓ Accepted photos are uploaded to Azure Blob Storage
- ✓ Six components return labels for visible image characteristics
- ✓ Text synthesis adds TCM educational context
- ✓ The architecture has not been independently clinically validated
Content cluster context
AI Tongue Analysis Platform
Trust-building content explaining the MyZenCheck system, benchmark methodology, and the role of AI-assisted TCM wellness screening.
Read next
AI Tongue Analysis Accuracy and Limitations: An Evidence Guide
What AI tongue-image systems can measure, why confidence is not diagnostic accuracy, how photo quality affects output, and what MyZenCheck has not clinically validated.
What MyZenCheck Can and Cannot Tell You
A clear guide to MyZenCheck's safety boundaries: what AI-assisted tongue analysis can support, what it cannot diagnose, and when to seek professional care.
Tongue App Privacy and Data: What to Check Before You Upload
A practical privacy checklist for AI tongue apps: what data is collected, how images are stored, retention, consent, and user controls.
Try AI-Assisted Tongue Analysis
Get educational TCM wellness insights based on visible tongue pattern clues
Try a Free Wellness Tongue Check