ingesting-clinical-documents

You can OCR scanned clinical documents using Tesseract or PaddleOCR to extract text and word-level source locations. This process preserves layout and maps detected text spans back to source pages and image bounding boxes for pixel-level review before de-identification.

The best way to redact direct identifiers from CSV or TSV exports is to classify columns as direct identifiers, quasi-identifiers, or safe fields, then apply masking, hashing, dropping, date shifting, or free-text redaction to ensure privacy-ready text.

Yes, you can convert C-CDA XML into normalized text through CDA flattening. This process transforms structured clinical documents into privacy-ready text suitable for downstream analysis without sending patient data to external services.

Yes, you need to convert unsupported PDF and DOCX files to images first. The OCR intake process requires image files to extract text using Tesseract or PaddleOCR backend for clinical document processing.

Yes, on-device AI clinical document processing works entirely without external services. Tesseract or PaddleOCR handle OCR locally, and the OpenMed multimodal extra enables privacy-preserving de-identification and entity extraction on-device.

Use column-aware redaction when processing structured CSV or TSV exports, classifying columns as direct identifiers, quasi-identifiers, or safe fields. Apply free-text redaction when dealing with unstructured clinical text or notes containing embedded PHI.