Radiology · Chest CT quantification and reporting support
AI-Rad Companion Chest CT
Siemens Healthineers
The reference example of big-vendor imaging AI: an eight-clearance FDA chain, a CE mark and a PMDA approval put automated lung, aorta, heart and spine measurements on every chest CT — quantification cleared as image processing, not CAD, with independent evidence that is candid about where the automation fails.
Performance Metrics
Clinical Evidence
No randomised trial of any AI-Rad Companion module exists — the evidence base is retrospective validation, reader studies and repeatability work, much of it independent and some of it usefully critical. The strongest standalone result comes from LMU Munich (Quantitative Imaging in Medicine and Surgery 2021): across 105 emergency whole-body CTs, the platform's per-nodule sensitivity for lung lesions was 96.7% against 90.0% for the original unaided reports, at 0.74 false positives per examination. In complex lung disease (Journal of Thoracic Imaging 2022, Medical University of South Carolina) per-nodule sensitivity fell to 67.7% — comparable to an expert reader — while patient-level classification reached 96.1%, the AI recovered 8.4% of nodules otherwise missed, and expert review time dropped from 2:44 to 0:36 minutes. The measurement modules have been validated principally by the University of Groningen's ImaLife programme. AI and manual thoracic-aortic diameters did not differ (32.7 versus 32.7 mm, European Journal of Radiology 2023), with AI–human discrepancy within inter-reader variability — but segmentation failed outright in 4.6% of cases. An independent UK evaluation on routine chest CT (British Journal of Radiology 2023, Royal United Hospital Bath) found good-to-excellent agreement (ICC 0.76–0.92) yet a full AI report in only 89% of 436 examinations. Repeatability on 3–4-month repeat low-dose scans is excellent (ICC above 0.9) for vertebral and aortic biomarkers, with coronary calcium volume the exception at 28.5% between-scan variability (European Radiology 2025 — a Siemens-funded Groningen study). The critical findings deserve equal weight. In participants with at least moderate emphysema the nodule-detection false-positive rate more than doubled, 0.51 versus 0.22 per scan (European Radiology Experimental 2024). For the Europe-only Chest X-ray module, an independent Bochum evaluation of 499 radiographs (Scientific Reports 2023) found sensitivity for lung lesions well above the written report (0.83 versus 0.52) but more false detections and lower sensitivity for pleural effusion (0.74 versus 0.88). The multicentre reader study reporting improved nodule detection with AI assistance (JAMA Network Open 2021) lists nine Siemens Healthineers employees among its authors and should be read as vendor-affiliated evidence; several Groningen authors also disclose Siemens grants or honoraria, which this catalogue records rather than adjudicates.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Rueckel J, Sperl JI, Kaestle S, et al. (LMU Munich; mixed authorship — Siemens co-authors) | RetrospectiveRetrospective | 105 | 96.7% per-nodule (standalone, lung lesions); 92.9% per-patient | — | — | Quantitative Imaging in Medicine and Surgery, 2021; emergency whole-body CT — AI surfaced secondary thoracic findings missed by original reports at 0.74 false positives per examination |
| Abadia AF, Yacoub B, Stringer N, et al. (Medical University of South Carolina; Siemens research-collaboration group) | RetrospectiveRetrospective | 143 | 67.7% per-nodule; 96.1% patient-level | 82.5% (40 nodule-free controls) | — | Journal of Thoracic Imaging, 2022; noninferiority vs expert reader in complex lung disease — AI recovered 8.4% of otherwise-missed nodules and cut expert read time from 2:44 to 0:36 min |
| Hamelink I, de Heide EEJ, Pelgrim GJ, et al. (University of Groningen / ImaLife; PI discloses Siemens grants) | RetrospectiveRetrospective | 240 | — | — | — | European Journal of Radiology, 2023; AI vs manual thoracic-aortic diameter 32.7 vs 32.7 mm (p=0.70), discrepancy within inter-reader variability — segmentation failed in 4.6% of cases |
| Graby J, Harris M, Jones C, et al. (Royal United Hospital Bath / University of Bath; independent) | RetrospectiveRetrospective | 436 | — | — | — | British Journal of Radiology, 2023; routine chest CT across three cohorts — manual-vs-AI agreement ICC 0.76–0.92, but a full AI report in only 89% of examinations |
| Hamelink I, van Tuinen M, Kwee TC, et al. (University of Groningen / ImaLife; Siemens PUSH-grant funded) | RetrospectiveRetrospective | 189 | — | — | — | European Radiology, 2025; repeatability on 3–4-month repeat low-dose CT — ICC >0.9 for vertebral, aortic and cardiac biomarkers; coronary calcium volume 28.5% between-scan variability |
| Sourlos N, Pelgrim GJ, Wisselink HJ, et al. (Groningen / Radboud; independent — critical finding) | RetrospectiveRetrospective | 121 | 0.68 in ≥moderate emphysema vs 0.71 without | — | — | European Radiology Experimental, 2024; false positives per scan 0.51 vs 0.22 (p=0.028) — nodule-detection specificity degrades in emphysema |
| Niehoff JH, Kalaitzidis J, Kroeger JR, et al. (Ruhr University Bochum; independent — Chest X-ray module) | RetrospectiveRetrospective | 499 | 0.83 lung lesions (report: 0.52); 0.74 pleural effusion (report: 0.88) | — | — | Scientific Reports, 2023; AI vs written report against two-radiologist + CT consensus — higher sensitivity for lesions, consolidation and atelectasis at the cost of more false detections; Europe-only module |
| Homayounieh F, et al. (MGH / LMU; nine Siemens Healthineers employees among authors — vendor-affiliated) | RetrospectiveRetrospective | 100 | — | — | — | JAMA Network Open, Dec 2021; multicentre reader study — AI assistance improved pulmonary-nodule detection on radiographs across difficulty levels and reader experience |
Clinical Pulse
Verified clinician sentiment
Verified clinician reviews launching soon · Apply to contribute →
Reviews are signed by clinicians verified against their regulator's public register.
Inside the algorithm
Editorial featureHow AI-Rad Companion reads a chest CT.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
Chest CTs flow through the teamplay cloud — from any vendor's scanner.
Studies route as series from the to Siemens' teamplay digital health platform, with an edge-deployment option recorded in the 2023 Cardiovascular clearance (K222360). Input is vendor-neutral: the software processes chest CTs from other manufacturers' scanners, not only Siemens systems. Analysis runs in the background alongside the standard read; source images are not modified.
Input
Chest CT DICOM
Inference
Background, per-examination
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
K222360
Class II
Safety Record
No recalls, field safety notices or MAUDE reports attributable to AI-Rad Companion were found in indexed public sources as of July 2026 — read as "none found", not an exhaustive audit, since the FDA recall and MAUDE databases could not be queried directly from this environment; Siemens safety actions that do surface (a 2025 MRI field safety notice, a Class 1 MRI recall) concern other Siemens products. The residual risks are those of any quantification layer: silent segmentation failure (4.6–11% of cases in the aorta studies yielded no or partial results), a false-positive burden that doubles in emphysematous lungs, and automation bias if measurements are accepted without inspection of the underlying contours.
Intended Use & Indications
AI-Rad Companion Chest CT is the marketed bundle of Siemens Healthineers' AI-Rad Companion platform engine and its body-system extensions for the thorax. Per the FDA 510(k) record, the Pulmonary extension is "image post-processing software that uses CT DICOM data to support clinicians in the evaluation and assessment of lung diseases" — segmenting the lungs and lobes and, in later versions, computing per-lobe opacity masks; the Cardiovascular extension segments the heart and aorta, quantifies total coronary calcium volume and measures aortic diameters at guideline landmarks; the Musculoskeletal extension segments and labels vertebrae and measures their heights and mean density. Results are delivered to the reading environment as DICOM structured reports and overlay series; the interpreting radiologist remains the decision-maker throughout.