Aivory AI for diagnostic radiology: CT and MRI analysis

Automated detection of lung and heart pathologies on chest radiographs. Cardiothoracic index, Moore index, heart volume — calculated in 3 seconds.

Learn more

AIVORY AI · NOSE

Skull and paranasal sinus X-ray

Recognition of pathologies on skull radiographs: fractures, sinusitis, foreign bodies, developmental anomalies. Highlighted areas of interest and a description for the report.

Learn more

AIVORY AI · PROSTATE

Prostate MRI · CDSS

A clinical decision support system (CDSS) for early diagnosis of prostate cancer from mpMRI. PI-RADS probability map, lesion segmentation.

Learn more

Module 01 · Aivory AI Chest

Chest X-ray — 14 pathologies and 8 indices

A convolutional neural network trained on Russian and international radiograph datasets. It analyzes frontal and lateral views, generates a DICOM SR and embeds an overlay directly into the PACS study.

14 classes of lung pathology. Focal opacities, infiltrates, atelectasis, emphysema, pleural effusion, pneumothorax, calcifications, signs of tuberculosis.
Cardiometry across 8 indices. Cardiothoracic index, Moore index, right atrial ratio, aortopulmonary index, vascular pedicle height, silhouette area, heart volume.
Screening and worklist sorting. Suspicious images move to the top of the RIS worklist; normal ones are sent to a fast track.
DICOM SR + Secondary Capture. A structured report with bounding-box coordinates and confidence scores goes to the PACS; the overlay is saved as a new series.
Inference time — 3 s. On an NVIDIA T4 / A10 GPU. Without a GPU — 12 s on a modern CPU. Analysis starts automatically when DICOM data is received.
On-premises, inside a protected network. The model runs on the healthcare facility’s server. Patient personal data and images never leave the perimeter; training is federated.

Aivory AI Chest · v4.2 · GPU

CR · chest · frontal view

patient F-58 · STUDY-2026-04412 · inference 2.8 s

Nodule S3 · 16×12 mm · 96%

Incr. lung markings · 78%

Cardiomegaly · CTI 0.58

DICOM SR · 1.2.840.10008.5.1.4.1.1.88.33 sent to PACS

Focal opacity, right lung S396%

Increased lung markings, left78%

Cardiomegaly (CTI > 0.5)94%

Pleural effusion / pneumothoraxnot detected

Cardiothoracic index

0.58 ↑ normal ≤0.50

Moore index

28% normal

Right atrial ratio

0.41

Vascular pedicle height

7.2 cm

Heart volume (frontal silhouette)

812 cm³ ↑ 720

Aivory AI Nose · v2.6 · skull / sinus

CR · skull · occipitomental (Waters)

patient M-34 · STUDY-2026-04471 · inference 2.1 s

Maxillary sinus opacity · 84%

Normal aeration · 97%

Fracture line · 91%

view: Waters · auto-detected SR · ACK 200

Left maxillary sinus · opacity84%

Frontal bone fracture line (right side)91%

Ethmoid labyrinthnormal

Foreign bodiesnot detected

Module 02 · Aivory AI Nose

Skull and paranasal sinuses — in two seconds

Recognition of pathologies on head radiographs: fractures of the skull vault and base, sinusitis, cysts, foreign bodies. Supported views: Waters, Caldwell, and nasal bones in lateral and axial projections.

Automatic view detection. The model recognizes on its own whether it is looking at a Waters, Caldwell or lateral nasal bone view, and selects the corresponding hypothesis.
Sinusitis, cysts, fractures. Opacification of the maxillary, frontal and ethmoid sinuses, cysts, foreign bodies, fracture lines of the nasal bones and skull vault.
Draft report for the radiologist. The AI generates a conclusion paragraph with references to the detected areas — the physician only edits and signs it.
Emergency department triage. If a fracture is suspected, the study automatically goes to the CITO (urgent) queue and is highlighted in the RIS worklist.
Explanatory heatmap (Grad-CAM). You can see exactly what the model is looking at — the physician accepts or rejects each finding.
Compatible with any CR. Digital X-ray units from any manufacturer — the model is robust to exposure, rotation and cropping.

Module 03 · Aivory AI Prostate

CDSS for early prostate cancer diagnosis

Clinical decision support based on multiparametric prostate MRI. Gland segmentation, detection of suspicious lesions, calculation of PI-RADS probabilities, registration of T2 / DWI / ADC series.

Gland and zone segmentation. Transition and peripheral zones, seminal vesicles, neurovascular bundles. Gland volume is calculated automatically.
PI-RADS probability map. Every voxel of the gland receives a PI-RADS v2.1 score of 1–5; lesions scored ≥3 are highlighted on T2, DWI and ADC simultaneously.
Series registration and fusion. T2, DWI b=1500, ADC and DCE are brought into a common geometry — the physician sees the same lesion on all series.
Structured PI-RADS report. A ready-made disease-specific template — coordinates, size, PSA density, biopsy recommendations. Signed with a qualified electronic signature (QES).
Biopsy referral routing. For PI-RADS 4–5 lesions, the patient is automatically added to the urologist’s queue via integration with the HIS (hospital information system).
Follow-up monitoring. On a repeat mpMRI after 6 / 12 months, the model compares lesions with the previous study and calculates volume growth.
Gland dimensions and volume. Calculated automatically from the axial and sagittal series: dimension accuracy of at least 0.86 and 0.87, volume accuracy with series matching — at least 0.92.
Pre-analysis check. A built-in model classifies studies by patient sex: studies that do not meet the criterion are not passed to the algorithm.

Aivory AI Prostate · v3.1 · mpMRI

Prostate MRI · PI-RADS heatmap

patient M-67 · STUDY-2026-04598 · PSA 8.4 · volume 48 cm³

T2 · TSE

DWI b=1500

ADC map

Highest PI-RADS

5

PZ-pl · left lobe · 14×11 mm

Total lesions

3

PI-RADS ≥ 3 · highlighted

Gland volume

48 cm³

PSA density 0.18 ng/mL/cm³

Recommendation

biopsy

ultrasound-guided fusion

          PI-RADS 3 · equivocal
          PI-RADS 4 · suspicious
          PI-RADS 5 · high probability
        

Model catalog

12 modules and an open API

The Aivory AI family covers the main modalities and screening target conditions. Third-party partner algorithms are connected through the DICOM router in one day.

PROD

Aivory AI Lung-CT

CT · low-dose screening

Detection of lung nodules on low-dose CT, volume calculation per Lung-RADS, change tracking on follow-up studies.

ICD C34

PROD

Aivory AI Mammo

mammography · BI-RADS

Digital mammography analysis: microcalcifications, masses, breast density per BI-RADS A–D.

ICD C50

PROD

Aivory AI Bone

limb X-ray · fractures

Detection of fractures of the long bones, wrist and hand. CITO triage for the emergency department.

ICD S52/S62

PROD

Aivory AI Stroke

head CT · ischemia / hemorrhage

Segmentation of ischemic and hemorrhagic stroke lesions on non-contrast CT. ASPECTS scoring and a CITO alert.

ICD I63/I61

BETA

Aivory AI Spine

spine CT · MRI

Vertebral segmentation, vertebral body height measurement, detection of compression fractures and disc herniations.

ICD M48/M51

BETA

Aivory AI Liver

contrast-enhanced abdominal CT

Liver and vessel segmentation, LI-RADS volumetry, detection of focal lesions and cysts.

ICD K70-K77

PARTNER

Cardio-AI · ECG + echo

partner · SberMedAI

ECG signal analysis and ejection fraction calculation from echocardiography. Connects to Aivory via the DICOM router.

partner· SaaS license· on-prem

PARTNER

Pathology-AI · WSI

partner · UNIM

Analysis of digitized histology slides (WSI), microstructure segmentation, cervical screening.

partner· on-prem· LIS bridge

BETA

Aivory AI Thyroid

thyroid ultrasound

TI-RADS classification of thyroid nodules on ultrasound images in real time.

ICD E04

Open API for algorithms

Connect your own models or third-party vendor solutions via the DICOM router and REST API. MONAI Deploy, NVIDIA Clara, ONNX and Triton Inference Server are supported.

DICOM C-STORE / WADO-RS REST · JSON · DICOM SR MONAI Deploy NVIDIA Triton ONNX Runtime HL7 FHIR · ImagingStudy Webhook callbacks

# Register an AI module
POST /api/v1/ai/registry
{
  "name": "my-pneumo-detector",
  "modality": "CR",
  "endpoint": "https://my-ai/infer",
  "input": "DICOM",
  "output": "DICOM-SR + Heatmap",
  "trigger": "on-receive",
  "timeout_ms": 30000
}
# PACS pushes every CR study
# the SR report is returned to the study automatically

Where it fits

Use cases for AI

One engine — five radiology scenarios: mass screening, emergency department triage, second opinion, preventive check-ups, teleradiology centers of expertise.

Mass screening

Fluorography, mammography, low-dose CT — high-volume study flows where the AI processes 100% of incoming studies and moves suspicious images to the front of the queue.

Triage · CITO

Stroke CT, polytrauma, fractures — the AI flags critical studies within seconds and escalates them to the on-call radiologist with a push notification.

Second opinion

AI as a “second radiologist” — helps avoid missing small nodules, calcifications and early lesions. All findings are shown as an overlay in the DICOM viewer.

Preventive check-ups

Chest X-ray and mammography for all employees / the enrolled population — the AI batch-reports “normal” studies, and the radiologist focuses on pathology.

CMIA teleradiology

A regional archive (CMIA, Central Medical Image Archive) handling thousands of studies per day — the AI sorts the flow so the center of expertise can report on what matters most.

Teaching residents

The AI attention map and confidence labels are an excellent hint for a young specialist: where the model is uncertain and why.

Integrations

Fits into your existing PACS / RIS

The AI listens on the DICOM receiver and returns results as SR and Secondary Capture; the overlay appears in all compatible viewers. Works with any PACS, not just ArchiMed.

ArchiMed PACS / Workstationnative integration

Third-party PACSDICOM C-STORE · WADO-RS

DICOM SRSOP 1.2.840.10008.5.1.4.1.1.88.33

Secondary Captureoverlay in a new series

RIS worklistpriority + CITO flags

HIS / EMIAS (Moscow’s unified medical information system)REST API · HL7 FHIR

EGISZ · electronic medical documentsSEMD (structured documents) submission to the Unified State Health Information System

Regional unified RISimage routing

MONAI DeployNVIDIA pipelines

ONNX · Tritonyour own models

QES · CryptoProphysician signs the AI report

Open SDKPython · C# · REST

Technical specifications

AI results — in the physician’s worklist

ArchiMed links the RIS, PACS and AI services — both its own and third-party. The analysis result is shown in the study table and is searchable.

AI request

No data copying

The request to the AI service passes the location of the study data, including the PACS address
Multi-frame DICOM studies are split into frames in the correct order
The patient’s PACS identifier is stored in the RIS and used for search

Result

Structured response

A parser converts the analysis result into a structured form
Detected pathologies are stored in a study attribute
The result is matched to the specific study and its processing flag

Worklist

“AI result” column

An AI result column in the study table, with a filter by it in advanced search
“Not processed by AI” and “no pathologies detected” are different states: a study without successful processing is not considered normal

Regulatory compliance

Russian medical AI software with local inference. Aivory AI modules are undergoing medical device registration with Roszdravnadzor.

Register

Unified Register of Russian Software

Rights holder — ArtVision LLC

Roszdravnadzor

Medical device registration

in progress

FSTEC

Certificate of conformity

personal data protection (FSTEC — Federal Service for Technical and Export Control of Russia)

Standards

DICOM SR · HL7 FHIR

results to PACS and RIS

Auto

analysis starts
when DICOM is received

SR

results go to PACS
as DICOM SR

12

modules in the catalog
+ an open API for partners

3 s

inference time
per CR study

Artificial intelligence for diagnostic radiology

Chest X-ray — 14 pathologies and 8 indices

Skull and paranasal sinuses — in two seconds

CDSS for early prostate cancer diagnosis

12 modules and an open API

Open API for algorithms

Use cases for AI

Fits into your existing PACS / RIS

AI results — in the physician’s worklist

No data copying

Structured response

“AI result” column

Regulatory compliance

Let’s run an AI pilot
on your studies

Artificial intelligence for diagnostic radiology

Chest X-ray — 14 pathologies and 8 indices

Skull and paranasal sinuses — in two seconds

CDSS for early prostate cancer diagnosis

12 modules and an open API

Open API for algorithms

Use cases for AI

Fits into your existing PACS / RIS

AI results — in the physician’s worklist

No data copying

Structured response

“AI result” column

Regulatory compliance

Let’s run an AI piloton your studies

Let’s run an AI pilot
on your studies