Pipeline walkthrough

From scan arrival to pre-annotated viewer in under one second.

A walkthrough of Neurmorph's annotation pipeline — what happens between PACS receive and radiologist open.

<1s
end-to-end annotation latency
4
pipeline stages, fully automated
0
changes to radiologist workflow

The 4-stage pipeline

01
DICOM Receive
Study arrives via C-STORE from your PACS modality worklist. Neurmorph registers the series UID, checks for priors, and queues for processing.
0ms – 50ms
02
Series Analysis
AI model evaluates slice stack. Identifies candidate findings by density, morphology, and spatial context. For chest CT: nodule candidates, airspace disease, pleural effusion, mediastinal widening.
200ms – 650ms
03
Measurement + Classification
Each finding measured in axial, coronal, sagittal planes. Density (HU) sampled. Prior series loaded and registered; growth-rate delta calculated. Fleischner / Lung-RADS estimate generated.
80ms – 120ms per finding
04
DICOM SR Delivery
Structured Report written back to PACS as DICOM SR object. Radiologist's viewer displays pre-loaded annotations on the correct slices.
20ms – 40ms

The radiologist is always in control. The AI is a first reader, not a final reader.

Every annotation is a suggestion, not a decision. The radiologist can accept (one click), adjust dimensions (drag measurement handles), dismiss (mark not-a-finding), or add findings Neurmorph didn't flag. Nothing is written to the report until the radiologist confirms it.

Every override is logged. That feedback is how the model calibrates to your specific scanner population and patient demographics — not a generic training set.

Typical deployment: one 2-hour IT session.

1
Confirm PACS DICOM connectivity (AE title, IP, port)
2
Neurmorph test node receives 5 deidentified studies
3
Latency baseline measured on your network segment
4
Live queue enabled on chest CT modality
No EHR integration required. No HIPAA BAA delay — we sign same day.

Ready to see it on your studies?

Send us 10 deidentified chest CTs and we'll return them pre-annotated within 24 hours.