Prioritization of critical cases
It reads the clinical data — not the image — and reorders the worklist in milliseconds, pushing what is urgent to the top before the next study is opened.
- Real time
The AI that respects the radiologist
Artificial intelligence in radiology
It acts at three moments — before the read, during dictation and after the signature — and at none of them does it block or diagnose. The clinical decision is always the radiologist’s.
AI at the edges of the workflow
Signals it readsSupport it returns
Signals it reads
Clinical dataHL7 · FHIR
Report textAs it is written
Viewer findingsDICOM · measurements
AI engineAssists · never replaces
Support it returns
Prioritized worklistCritical cases on top
Quality controlBefore delivery
Real-time alertsLaterality · size
Our philosophy
The AI that respects the radiologist
Everyone in the market promises "radiology AI". We start from a different premise: AI is not there to decide the diagnosis, it is there to take friction away from whoever decides it. Four principles govern every model we put into production.
It assists, it does not replace
No finding is closed without the radiologist. The AI suggests, prioritizes and verifies; the signature and the clinical responsibility stay human.
At the edges of the workflow
It never blocks the read and never diagnoses. It acts before — prioritizing the queue —, during — flagging without slowing anyone down — and after the signature — auditing reports in batches.
Explainable and traceable
Every alert states why it fired and against which piece of data. Everything is audited, so each decision can be reviewed with clinical judgment.
On your data, under your governance
It runs inside your own deployment — cloud, on-premise or hybrid — with encryption, role-based access control and compliance with Colombia’s Law 1581 of 2012 on data protection.
Report quality control
It reviews signed reports in batches and flags anatomical and laterality inconsistencies for review, without slowing down live reading.
- Post-signature
Inconsistency detection
It flags instantly when a laterality or a measurement in the text does not match what is marked in the viewer, so it can be fixed before signing.
- While writing
Operational intelligence
The AI that respects the radiologist
At NOVA Imaging, the AI never blocks the read and never diagnoses for the radiologist. It acts at three moments: before the read it prioritizes the queue, during dictation it flags without slowing anyone down, and after the signature it audits in batches. The clinical decision is always the human’s.
Before the read · in real time
Prioritization of critical cases
NOVA Imaging reads the clinical data —not the images— and reorders the worklist before the radiologist opens the next study. What is urgent rises to the top on its own.
The AI can:
- Read the clinical data (HL7 · FHIR), never the image in order to diagnose
- Reorder the worklist in milliseconds, in the background
- Prioritize critical cases by severity and by your rules
intelligent worklist · AI prioritization
Clinical data (not the image)
HL7 ADT · admitted with chest pain
FHIR · elevated troponin (3.1 ng/mL)
Order · Chest CT · "rule out PE"
History · D-dimer > 2000
ER triage · ESI-2 priority
Score 96/100 → probable PE. Raise to STAT.
Worklist · reorderedlive
STAT
García R., M.
CT · Chest
96
TODAY
Vega A., R.
US · Carotid
41
TODAY
Ramírez K., L.
XR · Chest
34
48h
Núñez T., S.
MR · Lumbar
18
24h
Soto V., M.
MR · Knee
12
Median latency: 180 msprocessing queue
After the signature · in batches
Report quality control
This layer is different from the inline alerts: it runs after the signature, in batches, over the already closed report. It flags serious anatomical and laterality inconsistencies for review the next day, without touching the live workflow.
The AI can:
- Review signed reports in batches, after the live workflow
- Contrast laterality and anatomy in the text against what was marked in the study
- Flag only what is serious, without slowing the radiologist’s reading
quality control · batch audit
Signed reports · day247 today
Batch control
08091011121314151617181920212223
Batch findings · review next day3 of 247 · 1.2%
R-2289Laterality · "left" vs R markeralta
R-2304Anatomical mismatch · organ referencedalta
R-2317Incomplete template · findingsmedia
False positives: < 0.3%overnight batch · 02:00
During the read
An assistant looking over your shoulder, without getting in the way
Real-time inconsistency detection
While the radiologist writes, the AI contrasts the text against the measurements and markers in the viewer and highlights discrepancies instantly — not after the signature.
- Laterality check (left/right) against the image markers
- Measurements in the text contrasted against those in the viewer
- Fewer reworks and fewer addenda after signing
report editor · AI
Findings
Focal lesion in the left kidney,
of cystic appearance, measuring 42 mm
with well-defined borders.
AI detectionLive
Laterality
“Left” does not match the R marker on the image.
Classification
Renal cystic lesion with no Bosniak category assigned.
Trust and governance
AI with clinical governance, not a black box
In healthcare, trust is not optional. That is why every AI capability is explainable, runs on your own data and leaves an auditable trail of every decision.
Protected data
Encryption in transit and at rest, role-based access control and audit traceability, aligned with Colombia’s Law 1581 of 2012 (Habeas Data).
The human decides
The AI does not issue diagnoses on its own. It suggests and verifies; the radiologist confirms, adjusts or discards every result.
Deployment on your terms
The AI runs inside your NOVA Imaging platform: fully in the cloud, on-premise in your data center, or hybrid, as your operation requires.
Continuous validation
We monitor the false positive rate (< 0.3%) and validate models with pilot institutions before taking them to production.
Frequently asked questions about AI in radiology
Does NOVA Imaging’s artificial intelligence replace the radiologist?
No. The AI prioritizes, audits and verifies, but it never issues a diagnosis on its own: the read, the signature and the clinical responsibility always stay with the radiologist.
Does the AI read images in order to diagnose?
Not to diagnose. For prioritization, the AI reads the clinical data (HL7/FHIR), not the pixels. Its support lies in ordering the queue, checking the consistency of the report and auditing its quality; interpretation and the decision are always the specialist’s.
Does the AI interrupt reading in real time?
No. It works at the edges of the workflow: before the read, prioritizing the queue, and after the signature, auditing reports in batches. The only live signal is the consistency alerts while the radiologist writes, which they can act on or ignore.
Where is patient data processed?
Inside your own NOVA Imaging deployment — cloud, on-premise or hybrid — with encryption, role-based access control and traceability, in compliance with Colombia’s Law 1581 of 2012 on personal health data.
How reliable is it? Does it raise a lot of false alarms?
We monitor the false positive rate below 0.3% and validate every model with pilot institutions before production. On top of that, every alert is explainable: it states why it fired and against which piece of data.
Do I need to change my RIS/PACS to use the AI?
No. The AI is part of the NOVA Imaging RIS/PACS platform and runs on open standards (HL7, FHIR, DICOM), built into the same workflow your radiologists already use.
Let’s talk about AI in your operation
We will show you, with data close to your own, how AI prioritization and quality control feel in a radiologist’s day.
Message us on WhatsApp