Back to work
BackendAI/ML
Medical-Imaging Anonymization Service
Healthcare / medical imaging
A microservice that automatically detects and redacts burned-in patient information from DICOM medical images using deep-learning text detection and OCR, plus 3D mesh extraction for visualization.
The Challenge
DICOM imaging files often contain burned-in patient identifiers that must be removed before images can be shared or reused — a slow, error-prone manual task.
What We Built
- An ASP.NET Core 8 REST microservice that parses DICOM with fo-dicom.
- ONNX deep-learning text detection plus Tesseract OCR to locate and mask burned-in PII on images.
- 3D mesh extraction and glTF optimization for downstream visualization, with presigned R2 download URLs.
Highlights
- Automated PII redaction on DICOM imagery
- ML text detection (ONNX) + OCR pipeline
- 3D mesh extraction with glTF optimization
- Secure presigned delivery