Platform
How Datamonk reads, fixes and delivers imaging data
A technical overview of the pipeline: what gets analyzed, how fixes are decided, who approves them, and how data is deployed and protected.
Predictable results, not black-box AI
Five principles govern every correction the platform makes.
Deterministic first
Most studies resolve against curated medical knowledge: known procedures, naming conventions and codes. AI is used only where certainty runs out.
Edge cases in full context
The hard studies are analyzed with everything available: all metadata across the hierarchy, cross-field consistency, and clinical context. From imports labeled "outside study" to copy-paste mistakes repeated across an archive.
Multilingual by design
Metadata is read across languages, vendors, abbreviations and local naming habits. One engine for every archive.
The image is the ground truth
Proposed fixes are checked against the pixel data itself: body part, laterality, contrast.
Clinical teams in control
Edge cases are grouped for clinical review. Approved fixes run automatically, every change is logged, and every learning makes the next study easier.
From input tokens to clinical context
Metadata isn't corrected field by field. Tokens from across the DICOM hierarchy are mapped to standardized concepts, then validated against each other — laterality against paired structures, body parts against their anatomic parents, protocol against modality.
One real study, before and after
Coverage across the DICOM hierarchy
100+ tags analyzed per study, at every level:
PATIENT
Patient ID · Name · Sex · and more
STUDY
Study Description · Procedure Code · Accession No. · and more
SERIES
Series Description · Body Part Examined · Laterality · and more
IMAGE
Anatomic Region · Modality · Contrast · and more
Output is normalized and SNOMED CT-coded, de-duplicated, and validated — delivered to any target: new PACS, cloud archive, VNA, or research environment.
Patient
Study
Series
Instance
Equipment
StudyDescription
97.8% resolved automatically
ProcedureCodeSequence
99.1% resolved automatically
AccessionNumber
100% resolved automatically
ReferringPhysician
96.4% resolved automatically
Deployment and security
Architecture
Dedicated tenant per hospital
Dedicated regions: EU and US
Runs on AWS; sources on-premise or cloud
Any PACS vendor and format, source and target
Test it on your own archive
Start small: a Data Quality Report analyzes 10,000 studies from your archive — any modality, any vendor — and delivers findings with sample corrected files, in 4–6 weeks. It begins with a 30-minute intro call.