Synnatrix

The challenge

Early oral cancer and potentially malignant lesions often look like ordinary white or red patches, ulcers or normal anatomy. Outcomes depend heavily on the stage at diagnosis, yet most dental clinics have no consistent screening step, and photos, health history and notes sit in different places.

What I built

• Two-stage AI pipeline: stage 1 detects candidate regions in an intraoral photo; stage 2 classifies each region's tissue micro-texture as normal, oral cancer, leukoplakia or lichen planus. A typical result takes about 6 seconds.
• Region-level results: every region is classified separately, with heatmaps, and false positives on normal anatomy are suppressed.
• Uncertainty built in: borderline findings are flagged as uncertain instead of being forced into a label.
• Clinical workflow: patient registration, a full digital health-history form (medications, allergies, risk factors), medical alerts, clinician notes and triage recommendations such as specialist referral or biopsy.
• SaaS subscriptions: Starter, Professional and Clinic plans with scan quotas, plus manual verification of each clinic before activation.
• Privacy by design: contact details and medical notes encrypted at rest; GPS and other photo metadata stripped on upload.
• Doctor portal, plus full legal pages: privacy, terms, renewal, refund, cookies and DMCA.

Positioning

A screening and triage aid, not a diagnosis: the final decision always rests with the clinician and histopathology.