What it does

Head CT reads the emergency head study as a whole. Upload a non-contrast head CT on its own, or together with a CT angiogram of the head and neck and a CT perfusion acquisition; the run recognises which series are present, picks the brain and bone reconstructions, and runs every section those series support.

The result answers the questions of the night shift: is there a bleed, in which compartment and how large; is there mass effect, midline shift or hydrocephalus; are there vault, skull-base or facial fractures. In a stroke context without a bleed it adds ASPECTS. With a CTA it adds large-vessel occlusion, intracranial aneurysms and the Circle of Willis; with a CTP, perfusion maps with core and hypoperfusion volumes. Trauma studies also receive a Marshall and Rotterdam CT class.

Intended use

Decision support for the first reader of a head CT: the resident, emergency physician or teleradiologist who must exclude a bleed before thrombolysis, decide on CTA and perfusion in a stroke code, or escalate a head injury to neurosurgery. A triage flag lets a worklist put critical studies first. Each finding comes with its mask or measurement.

Who it is for

Emergency physicians, stroke neurologists, neurosurgeons, radiology residents and teleradiology groups. Developers call the same run to add head CT triage and quantification to a worklist, stroke pathway or PACS.

Accepted input

• CT series: CT Image Storage, Enhanced CT or Legacy Converted Enhanced CT, uploaded as a zipped study, through DICOMweb STOW-RS or from a pre-signed URL.
• Head study: non-contrast CT in brain and bone kernels, optionally a head and neck CTA and a 4D perfusion series in the same upload. Localizers, dose screens and derived reformats are set aside when an original axial series exists.
• Prior head CT in prior_files, used by the comparison section.

Requirements

Requirement	Why
Image Position and Orientation (Patient), Pixel Spacing, Slice Thickness	Slices are ordered geometrically and gantry tilt is detected
Rescale Slope and Intercept	Bleed density and volumes are computed in Hounsfield units
Thin axial reconstruction where available	5 mm slices can hide small bleeds; thick studies are read and flagged
Acquisition Time on every perfusion time point	Deconvolution of the perfusion series needs the time axis
Arterial timing of the CTA	Venous contamination limits the occlusion and aneurysm reads and is flagged

A section whose input is missing is marked as not assessed: no ASPECTS when a bleed is present, no occlusion read without a CTA, no perfusion maps without a CTP.

Optional context

The clinical question (stroke, trauma, follow-up of a known bleed) and prior findings go in clinical_context. They decide whether ASPECTS or the Marshall and Rotterdam classes are reported, and they shape the guidance.

The result

One JSON result in sections with findings, measurements, volumes in millilitres and quality flags for motion, posterior-fossa beam hardening, thick slices and post-operative change. Masks reference the original slices.

Section	Content
Critical-finding triage	Urgency flag for bleed, mass effect, hydrocephalus, large infarct or fracture
Intracranial hemorrhage	Epidural, subdural, subarachnoid, intraparenchymal and intraventricular blood: 3D masks and volume per compartment
Midline shift	Septum pellucidum deviation in millimetres, effacement and herniation signs
Ventricles and hydrocephalus	Ventricle volumes, Evans index, temporal-horn dilatation and a hydrocephalus flag
Skull and facial fractures	Vault, skull-base and facial fractures on the bone kernel, depressed fracture, pneumocephalus
ASPECTS	Early ischaemic change in each region (C, L, IC, I, M1–M6) with side, and the 0–10 score
Large-vessel occlusion	Occlusion present or absent, site (ICA terminus, M1, proximal M2, basilar) and side
CT perfusion	CBF, CBV, MTT, TTP and Tmax maps; core and hypoperfusion volumes; mismatch ratio
Intracranial aneurysm	Each aneurysm with its vessel segment and maximal diameter
Circle of Willis segmentation	Multiclass vessel mask and centreline graph, including variants such as a fetal PCA
Comparison with prior	Haematoma expansion (absolute and relative), new bleeds, change in shift and ventricles
Draft report	Opt-in (options.report): an English narrative that interprets the result for the reader (findings, impression, limitations); every number is checked against the findings

Standards

• DICOM: SR TID 1500 Measurement Report for volumes and diameters; SEG for bleed, ventricle and vessel masks; Parametric Map objects for the perfusion maps; GSPS overlays, key images, Encapsulated PDF and FHIR DiagnosticReport.
• Scores: ASPECTS (Barber 2000); ICH volume with an ABC/2 cross-check and the imaging inputs of the ICH score; Evans index; Marshall and Rotterdam CT scores for traumatic brain injury; DEFUSE-3-style perfusion thresholds.

Thresholds

Midline shift above 5 mm and an Evans index above 0.3 are flagged. Perfusion core is tissue with rCBF below 30 %; hypoperfusion is Tmax above 6 s. Every flag is recomputed from its stored measurement, so the number and the flag always agree.

Limitations and failure modes

• Sections in version 1.0.0: hemorrhage and triage. Midline shift, ventricles, fractures, ASPECTS and the CTA/CTP sections are listed as not assessable until they join the run.
• Small bleeds on 5 mm slices: bleeds under about 5 mL on 5 mm reconstructions are found less often than larger ones; the high-sensitivity point (hemorrhage "not excluded") is there for the rule-out question.
• Subdural outlines are the least exact: large chronic collections are under-called in volume.
• Epidural bleeds are detected and outlined; their volume and thickness are not reported in this version.
• Infratentorial origin (an ICH score input) is not assessed yet.
• Contrast-enhanced series are not read for hemorrhage; a study with contrast series only is rejected.
• Post-operative heads, calcified falx and posterior-fossa beam hardening can mimic subarachnoid blood.