Dental AI Diagnostic Assistant & Clinical Tools
AI and Computer Vision Engineer at Smartory Labs
Role and Mission Overview
At Smartory Labs, clinical AI tools were engineered for automated dental radiograph diagnosis, at-home treatment follow-up, and daily oral hygiene monitoring using smartphone cameras.
Automated Dental Radiograph Diagnosis
The Challenge
Subtle early-stage cavities, hidden restorations, and overlapping anatomy in dental radiographs often present significant diagnostic challenges during rapid chairside evaluation.
Clinical Data Operations & Dataset Curation
Spearheaded end-to-end data gathering and multi-stage annotation protocols across 100,000+ radiography images in close collaboration with over 20 dental specialists, establishing high-quality ground truth benchmarks for detection, segmentation, and classification.
The Solution and System Architecture
A two-stage deep learning diagnostic pipeline was engineered: an initial model detects individual teeth and update their profile in the application, after which a secondary model crops each tooth to diagnose defects. Diagnostic flags are generated for subtle abnormalities, providing clinicians with an automated chairside second opinion.
Multi-Modality Radiographic Analysis
High-resolution clinical demonstrations across complementary radiographic modalities:
Bitewing Interproximal Defect Detection
Targeted bounding box localization of hidden interproximal caries, marginal bone level changes, and sub-surface restorations.
Full-Arch Anatomical Segmentation
Multi-class semantic segmentation identifying individual dental numbering, alveolar bone contours, and mandibular nerve canals across panoramic scans.
Smartphone Hygiene Monitoring and Patient Profiling
The Challenge
Post-treatment follow-up and preventative oral hygiene tracking between clinical appointments have traditionally been hindered by arbitrary camera angles, inconsistent lighting, and lack of continuous monitoring tools.
The Solution and System Architecture
A mobile computer vision pipeline was developed alongside a custom-designed physical smartphone holder to standardize camera alignment and illumination during at-home imaging. The system was designed for treatment follow-up and daily oral hygiene monitoring, allowing longitudinal patient profiles to be tracked and updated automatically.
AugmenTory Open-Source Polygon Library
The Challenge
Instance segmentation models require extensive data augmentation, yet traditional raster mask tools consume excessive memory and introduce substantial throughput bottlenecks during high-resolution dental imaging training.
The Solution and System Architecture
AugmenTory was conceived and released as an open-source Python library for native vector polygon transformations. By executing geometric calculations directly on coordinate vertices instead of pixel masks, spatial memory overhead was reduced to 1.2% while accelerating training batch generation by 1.7x.
Vector Polygon Augmentation & Crop Pipeline:
