How DexaVision Diagnoses Dental Caries in Under 60 Seconds Using Claude on AWS Bedrock
Manual dental diagnosis takes days, costs money, and remains inaccessible in rural areas. DexaVision replaces the traditional pipeline with a multi-model AI system that delivers ICDAS II clinical classifications from a single dental photo — in seconds, not weeks.
The Challenge
Dental diagnosis is slow, expensive, and out of reach for millions
Dental caries is the most prevalent chronic disease worldwide. Yet diagnosis still requires in-person visits, long wait times, and manual visual evaluation by a specialist. Patients in rural areas of Latin America often travel hours for a single appointment, only to wait days for results. Dentists spend valuable clinical hours on initial screening that could be automated.
Before DexaVision
Patient calls clinic
Waits 3-7 days for appointment
Travels to clinic (often hours away)
Sits in waiting room
Manual visual diagnosis
Results delivered days later
Average time to diagnosis: 3-7 days
After DexaVision
Upload dental photo from phone
AI validates image quality instantly
ICDAS II classification in 60 seconds
Consult with verified dentist in < 1 hour
Full clinical report generated
24/7 access from anywhere
Average time to diagnosis: Under 60 seconds
The Solution
Multi-Model AI Pipeline on AWS
DexaVision orchestrates four AI models through AWS Step Functions, running parallel per-image analysis. Each model handles a specialized stage of the diagnostic pipeline, with a dedicated AI arbiter as the final clinical validator.
Image Validation
Validation AIValidates uploaded images are dental radiographs or intraoral photos, filtering out non-clinical uploads before they enter the pipeline.
Jaw Identification
Vision AIIdentifies the jaw region (upper/lower, left/right quadrant) to provide anatomical context for tooth detection and diagnosis.
Tooth Detection
Detection AIReal-time object detection model localizes individual teeth within the image, creating bounding boxes for per-tooth analysis.
Caries Specialist
Specialist AIAnalyzes each detected tooth for caries indicators, producing preliminary ICDAS II classifications with confidence scores.
Final Arbitration
Arbiter AIThe final arbitrator cross-validates all findings, resolves conflicts between models, and produces the definitive clinical diagnosis.
Why a Dedicated AI Arbiter?
The arbiter's advanced reasoning capabilities make it uniquely suited for the final validation stage. It cross-validates findings from three specialist models, resolves conflicting classifications, weighs confidence scores, and produces a unified clinical diagnosis that adheres to the ICDAS II standard — all while maintaining the nuanced judgment required for healthcare applications.
Why Claude
Why Claude on AWS Bedrock
Clinical Accuracy
Advanced AI reasoning handles complex multi-finding arbitration, cross-referencing outputs from 3 specialist models to produce reliable ICDAS II classifications.
Managed Infrastructure
AWS Bedrock eliminates model hosting complexity. No GPU clusters to manage, no model versioning headaches — just pay-per-use inference at scale.
Multi-Model Orchestration
AWS Step Functions coordinate multiple specialized AI models seamlessly — running parallel per-image analysis with automatic retries and error handling.
Enterprise Security
KMS encryption at rest and in transit, HIPAA-compliant image storage on S3 with OAC-protected CloudFront, and 7-year data retention policies.
Impact
Results
Technology
Technical Stack
Frontend
Backend
AI / ML
Infrastructure
Ready to Build with AI?
CloudForge AI helps you design, build, and deploy AI-powered solutions on AWS. From multi-model pipelines to serverless architectures — we turn complex ideas into production-ready products.