← Back to work
Research Projects
3D CBCT Tooth Segmentation
Researched, trained and deployed multiple deep learning models for 3D dental imaging, from raw data processing to live inference.
- Role
- AI Engineer | AI Researcher
- Context
- PhD research side project (team of 5)
- Period
- Dec 2025 – Present
Overview
An application that automatically analyzes dental CBCT scans: from a single 3D volume, it separates each individual tooth and the pulp inside it, replacing manual slice-by-slice segmentation.
Supports diagnosis and treatment planning: root canal measurement, implant and orthodontic preparation, and per-tooth 3D model reconstruction.
Key achievements
Developed a web application for visualizing and inferring pulp segmentation and root canal length measurements.
Applied and validated in a real customer use case.
Responsibilities
- Reviewed scientific papers on 3D segmentation methodologies, with a focus on medical image processing.
- Processed the dataset and deployed it to S3-compatible storage with version management.
- Built Attention U-Net and V-Net models for the segmentation task, training them and running enhancement loops to improve accuracy.
- Deployed the model on a serverless instance and developed an application with step-by-step visualization, allowing users to perform tooth segmentation, pulp segmentation, and heuristic root canal length measurement.
- Stored models and datasets on an S3-compatible volume for cloud-based training.
- Monitored training with the open-source tool WandB.
- Analyzed model behavior on real cases to improve handling of edge cases.
- Adjusted the model architecture to reduce inference cost and application latency in line with customer requirements.
Tech stack
Languages & Libraries
PythonOpenCVSimpleITKTensorFlowPyTorchCUDA
AI research
U-NetV-NetAttention U-NetMamba SSM