What it does

Chest X-ray reads a chest radiograph study and puts urgent findings first. Upload a frontal film (PA or AP), with a lateral view and a prior study when you have them. The run checks projection and exposure, flags pneumothorax, malpositioned tubes, large effusions, consolidation and subdiaphragmatic free air, then returns the full finding list.

Local findings come with boxes on the image; pneumothorax, effusion and consolidation come with size or extent; every endotracheal tube, nasogastric tube, central venous catheter and Swan-Ganz catheter comes with its tip position and, for the endotracheal tube, the tip-to-carina distance in millimetres. The cardiothoracic ratio, a comparison with the prior film and two opportunistic risk estimates complete the read.

Intended use

Decision support for the doctor who reads the film, often without a radiologist: an ED physician at night, an ICU team checking lines on the morning round, a GP, or a TB screening programme. Each finding comes with its evidence, and the guidance points to the matching pathway (BTS pneumothorax management, CT for a nodule, molecular testing for a TB-positive screen).

Who it is for

Emergency, intensive-care and respiratory physicians, radiologists and residents, GPs and TB programme staff. Children are handled inside this model through age-aware routing; neonatal films belong to NICU Radiograph. Developers call the same run to add chest X-ray analysis to a PACS, worklist or screening app.

Accepted input

• DICOM radiographs: CR and DX (For Presentation or For Processing) objects; one frontal image is enough.
• Studies: PA or AP frontal, optional lateral, and an optional prior study for the comparison section.
• PNG or JPEG for development: accepted, but marked uncalibrated, so no millimetre values are returned.

Requirements

Requirement	Why
A frontal view (PA or AP)	Every section starts from the frontal film; the lateral and prior add to it
ImagerPixelSpacing or PixelSpacing	Needed for tube-to-carina distance, nodule diameter and pneumothorax rim distance in mm
ViewPosition and patient position	AP supine films overestimate heart size and hide anterior pneumothorax; the ratio is flagged as unreliable
PatientAge or birth date	Selects the paediatric path below 18 years and the TB screening age (15 years and over)

MONOCHROME1 images are inverted automatically; rotation, poor inspiration, clipping and skin folds are reported in the quality section.

Optional context

Age, sex and smoking history feed the risk sections. The indication or clinical question (for example "ARDS?" or "line check") in clinical_context focuses the guidance.

The result

One JSON result in sections, with device positions, calibration and projection quality as structured fields.

Section	Content
Triage	Normal or abnormal, plus urgent flags for worklist priority
Image quality	Projection (PA, AP, lateral), rotation, inspiration, penetration, clipping, collimation and artefacts
Findings	Multi-label probabilities for common chest findings, with boxes for local findings
Pneumothorax	Presence, side, mask, rim distance, tension signs and chest-drain presence
Lines and tubes	ETT, NGT, CVC and Swan-Ganz tips, carina landmark, ETT–carina distance and a normal, borderline or abnormal position
Pneumonia	Airspace opacity and consolidation masks with probability, lobe and laterality
Tuberculosis	TB abnormality score 0–100 against a locally calibrated threshold, with boxes on TB-consistent findings
Lung nodules	Nodule and mass boxes, diameter in mm, calcified or not, and a CT recommendation
Cardiothoracic ratio	Heart and thoracic widths from masks and the ratio on PA films
Pleural effusion	Side and size grade: costophrenic blunting, meniscus, or half the hemithorax or more
Pulmonary oedema	Severity 0–3 (none, vascular congestion, interstitial, alveolar); RALE score when ARDS is queried
Fractures	Rib, clavicle, scapula and vertebral-body fractures, acute or healed, with rib numbering
Anatomy segmentation	Masks of lungs, heart, clavicles, scapulae, mediastinum, aorta, diaphragm and spine
Comparison with prior	Each finding improving, stable or worsening; new and resolved findings; device moves
Cardiovascular risk	10-year cardiovascular risk and a radiographic biological age
Lung cancer risk	Long-term lung-cancer risk and a low-dose CT screening suggestion
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: Chest CAD SR (TID 4100) for findings, TID 1500 Measurement Report for CTR and ETT–carina distance, TID 4019 Algorithm Identification, SEG for lungs, heart, pneumothorax and devices, GSPS overlays, KOS key images and an Encapsulated PDF; display through the IHE AI Results (AIR) profile.
• Terminology: Fleischner Society glossary (a nodule is 3 cm or less, a mass is larger), RadLex and SNOMED CT.
• Guidelines: BTS pleural disease guideline (2023); IDSA/ATS 2019 community-acquired pneumonia guideline; WHO consolidated TB guidelines, Module 2, and the WHO TB CAD target product profile; Berlin ARDS definition.
• Risk comparators: ACC/AHA pooled cohort equations; USPSTF 2021 low-dose CT eligibility, with Lung-RADS applying on the follow-up CT.

Measurement notes

Pneumothorax size is classed by two conventions: a large pneumothorax has a rim of 2 cm or more at the level of the hilum (BTS) or 3 cm or more from apex to cupola (ACCP). The adult ETT tip target is about 5 ± 2 cm above the carina. A CTR above 0.5 on a PA film sets the cardiomegaly flag; AP films report the ratio but mark it unreliable.

Limitations and failure modes

• Sections in version 1.0.0: triage, findings, pneumothorax, pneumonia, lines and tubes, projection. Boxes for local findings, pneumothorax size and chest drains, tuberculosis, nodule size, the cardiothoracic ratio, effusion grade, oedema score, fractures with boxes and the comparison with a prior are listed as not assessable until they join the run.
• Adults only: the networks were trained on adult films; films of patients under 18 years are rejected.
• Small or anterior pneumothorax on supine AP films is found less often than an apical rim on an erect film, and the outline of a small pneumothorax can be partial; skin folds can mimic a pneumothorax edge.
• AP portable films enlarge the heart and mediastinum: cardiomegaly and a widened mediastinum read more often.
• Operating points were set on VinDr-CXR (one hospital network); on low-resolution or heavily processed exports the panel finds fewer subtle findings.
• Central line and nasogastric tips are given as a position category, not as a tip point; the endotracheal tube tip and the carina are points.
• One frontal film is read per study; the lateral film is not analysed yet.