DentalXNet
Research Portfolio Project
DentalXNet: AI-Powered
OPG Radiograph Analysis System
A deep learning framework using a hybrid CNN-Transformer model (RT-DETR) for automated identification and segmentation of dental treatments in panoramic OPG radiographs.
Dataset Size
2,235 Scans
Mendeley OPG dataset
Class Count
6 Treatments
Crowns, implants, fillings, etc.
Proposed Precision
85.3%
RT-DETR-L performance
mAP@0.5
71.3%
+17.3% vs YOLOv8 baseline
Platform Capabilities
Hybrid CNN-Transformer Core
RT-DETR combines rapid local convolution backbone feature maps with self-attention layers to model relational coordinates across the entire jaw.
Privacy-Aware Data Architecture
Patient identities are locked strictly inside browser LocalStorage/IndexedDB. Scan inferences are stateless, run in-memory, and immediately dumped.
Interactive Vision Tools
Dynamic client-side canvas bounding box toggles, a confidence slider filtering, and adjustable heatmaps calculated in real time.
Mendeley Dataset Distribution
Breakdown of annotated classes from the 2,235 radiographs, highlighting the imbalance challenges (such as fillings vs endodontic posts).
Crown
1,500
Dental Fillings
5,900
Endodontic Post
650
Root Canal Treated Tooth
1,400
Bridge
620
Implant
850