SAQR MED — trained model for early visual diagnostics

Why liver, first

A global oncology burden with a structural diagnostic gap

Liver cancer is one of the deadliest cancers worldwide — not because treatment is unavailable, but because most patients arrive too late. SAQR MED is built for that class of patient and that workflow, in any country where the diagnostic window is being missed.

866K

New cases per year (global)

Liver cancer generated 866,136 new cases and 758,725 deaths worldwide in 2022 — the 6th most-diagnosed cancer and the 3rd leading cause of cancer death.

Source: GLOBOCAN / IARC WHO 2022

3rd

Leading cause of cancer death

With a mortality-to-incidence ratio of 0.86, liver cancer kills almost as many people as it diagnoses — a signature of late-stage discovery, not of untreatable biology.

Source: Chinese Medical Journal 2024

240M

Chronic hepatitis B carriers

WHO estimates 240 million people are living with chronic HBV infection — the primary driver of hepatocellular carcinoma. A large, screenable at-risk population.

Source: WHO Hepatitis B fact sheet 2026

70%+

Cases in Asia and Africa

Almost 79% of the global liver-cancer burden falls on Asia and Africa. This is where late-stage presentation is most common — and where an early-window model has the highest clinical impact.

Source: Global epidemiology of liver cancer 2022

What it is

Not a cloud service. Not a certified device. A trained model you own.

SAQR MED is a shipped software product for early visual diagnostics. We arrive not with a subscription or a device on regulatory approval, but with a trained model, the right to use it, and a team that helps the customer put it to work.

01

A trained model

A CNN trained on ~8,700 real liver MRI studies. Certificate of developer rights. First release — liver; on the roadmap — heart, pancreas, kidneys.

02

Usage rights

The customer receives the model for permanent use — not a subscription seat. Weight files, configuration and documentation all sit inside the customer perimeter.

03

Team training

We train the customer's radiologists and IT engineers to operate the model, read risk maps and interpret results correctly. Not "give us access" — a full transfer of expertise.

04

Platform and support

We ship the platform that connects the model to the clinic's PACS/DICOM and radiologist workstations, operate it, and retrain the model as customer data accumulates.

How it works

From MRI study to risk map — on the clinic's own hardware

SAQR MED sits next to the clinic's existing PACS. All inference happens locally — data never leaves the perimeter. The model produces a risk map for the radiologist, not an automated verdict.

SAQR MED architecture — MRI and PACS data flow into the local SAQR MED CORE, then to radiologist and clinician workstations and a PDF report

01

Study intake

An MRI or CT study is sent from the scanner or PACS to the local SAQR MED server. Standard DICOM format. No manual data preparation.

02

Local inference

The CNN processes the slice series on a mini-PC next to the scanner. No cloud calls, no external APIs.

03

Risk map, not an alert

The result is not a binary "sick/well" alert but a visual map of attention regions on the study, ranked by probability of pathology.

04

Physician-in-the-loop decision

The radiologist reads the risk map alongside the study and issues the report themselves. The model does not replace the physician — it shortens search time and reduces the probability of a miss.

05

Longitudinal dynamics

Each new study is compared against the patient's prior studies. The model tracks changes in lesion size and structure between visits.

Inside the model

Why this works

Four engineering choices that separate SAQR MED from generic AI services and from research prototypes.

CNN, not a language model

The core is a convolutional neural network trained specifically on medical images. We do not take a general vision model and fine-tune it on a few hundred scans — the model is built from scratch for a specific anatomy.

Trained on real clinical data

~8,700 liver MRI studies from diagnostic clinics, regional hospitals, oncology centers and private practices. Not open datasets, not synthetic — real clinical material.

Risk map, not a verdict

The model highlights regions that deserve the radiologist's attention and ranks them by probability. The physician reads the study — the model simply narrows the search space.

Dynamics between studies

The model retains a patient's study series and highlights change between them — lesion size, density, morphology. Valuable on long clinical horizons.

SAQR MED interface — axial liver MRI slice with two marked lesions, prior-studies panel on the right, series gallery below — Example interface: axial liver MRI slice, two marked attention regions with coordinates and dimensions, patient's prior-studies panel.

Model roadmap

Four anatomical zones

We move deliberately from one anatomical zone to the next. Each model is trained separately, tested separately and delivered to the customer separately. No universal "model of everything".

Working MVP

Liver

First released version. Trained on ~8,700 MRI studies. Optimized for early hepatocellular lesions — a priority for countries with high HCC incidence.

In development

Heart

A model for the early detection of myocardial and coronary pathologies on MRI and CT angiography. Expected release next after the liver model.

On the roadmap

Pancreas

One of the hardest organs for early diagnostics. Focal changes are often missed until late stages. This model addresses exactly that gap.

On the roadmap

Kidneys

A model for focal kidney lesions and their dynamics. Closes the set of four priority anatomical zones.

Where we deliberately do not play

Product boundaries matter as much as product capabilities

We name honestly what SAQR MED is not. That protects the clinic from wrong expectations, and us from wrong deployments.

Not a medical device

SAQR MED is software that assists the radiologist. We do not position the product as a certified medical device and we do not replace the clinic's regulatory processes.

Not mass radiology

We do not target "tens of thousands of scans per day" routine studies. Our job is early detection of priority focal pathologies — not throughput.

Not a cloud solution

The model runs on a local mini-PC inside the clinic perimeter. We do not collect patient data on our servers and do not send scans to third countries.

Not a subscription model

SAQR MED is not sold as SaaS. The customer receives the model and the rights to it, and pays for delivery and support — not a "seat in a system".

Commercial model

We sell the model. Not a subscription.

SAQR MED is transferred to the customer as a trained model with usage rights, together with deployment infrastructure and a training program for the team. Pricing is structured per project — figures are discussed individually.

What is included	What the customer receives
Model	A trained CNN for the chosen anatomical zone with usage rights and full documentation.
Infrastructure	A local mini-PC with the model pre-installed, integrated with the clinic's existing PACS/DICOM system.
Team training	A training program for the customer's radiologists and IT engineers: interface work, risk-map interpretation, platform operations.
Support	12 months of baseline support after go-live: SLA, updates, consultations, complex-case reviews.
Pilot	Option to start with a limited pilot on one clinical site, with subsequent expansion based on the outcome.

Pricing is structured per project and discussed individually.

Request terms

Team

The model was trained under supervision from practicing clinicians

We do not disclose the names of the medical scientific director or the partner clinics — this is a condition of working with clinical data. Below are the aggregate credentials of the team responsible for the product's clinical quality.

Medical scientific director

48 years of clinical experience in diagnostic radiology
359 peer-reviewed publications
3 patents in medical imaging
Doctor of Medical Sciences, Professor

Name withheld — the scientific director works with several SAQR MED clinical partners, and their involvement is covered by confidentiality agreements. Available under NDA.

Clinical team

4 practicing radiologists
All hold the highest qualification category
All are PhD candidates in medical sciences
Combined experience — over 80 years of clinical MRI/CT work

The team participates in training-data annotation, model validation and complex clinical-case reviews.

Engineering team

ML engineers with dedicated medical-imaging experience
Platform developers, DevOps and PACS/DICOM integration engineers
SAQR AI product team owning the radiologist UX

A full product team — not consulting, not outsourcing.

Illustrative clinical scenario

Day 1 → 3 months → 6 months

An example of how the model follows a single patient over time. Country, clinic and patient details are generalized — the scenario reflects a typical workflow.

Three sequential MRI studies of the same patient's liver — baseline, +3 months, +6 months — with evolving lesion annotation

Day 1

Baseline MRI

Patient attends a routine liver exam. The model highlights a small focal area a radiologist might miss under case load. Follow-up is scheduled at 3 months.

+3 mo

Control MRI

The model compares the new study against baseline — the lesion has grown from ~8 mm to ~12 mm. An automated comparison map is generated. The physician initiates treatment.

+6 mo

Post-therapy MRI

On the third study, the model shows significant reduction — annotation moves to "marked shrinkage". Patient stabilizes under observation, therapy is working.

Deployment path

From first conversation to a running platform — 6 to 9 months

We break deployment into six measurable stages. Each ends with a concrete result — not with "exploring possibilities".

01

Infrastructure audit (2 weeks)

We assess the clinic's PACS system, hardware, data flows and country-specific regulatory requirements. Output — a technical scope.

02

Demonstration (1 week)

We run the model on a representative sample of the clinic's anonymized studies. The customer sees the model on their own data before signing a contract.

03

Contract and delivery (2–4 weeks)

Commercial terms, licensing rights and delivery scope are finalized. Mini-PC is shipped, model files are transferred.

04

Installation and integration (2–3 weeks)

Deployment on the clinical site, integration with PACS/DICOM, role-based access setup, test runs.

05

Team training (4–6 weeks)

Training program for the customer's radiologists and IT engineers. Real clinical cases, risk-map interpretation, platform operations.

06

Support (12 months)

Planned SLA, model updates, complex-case reviews, consultations. The outcome informs the decision to move to the next anatomical zone.

Compact SAQR MED mini-PC next to an MRI scanner in a diagnostic room — SAQR MED local hardware sits next to the diagnostic scanner — data does not leave the clinic perimeter.

Security and compliance

Designed for the medical regulatory landscape

We do not rely on "trust us" — the product architecture is built so that patient data physically does not leave the clinic.

Local operation

The model and patient data sit on a mini-PC inside the clinic perimeter. No cloud calls, no scans on SAQR AI servers.

Role-based access

Radiologists, treating physicians and IT engineers only see what their role requires. Full permission segregation in the interface.

Audit log

Every interaction with the model and its results is recorded: who opened a study, what finding the model returned, what decision the physician made.

HIPAA / GDPR / PDPA

The architecture is designed to be compatible with HIPAA (US), GDPR (EU) and regional data-protection frameworks. Specific compliance is contractual work with the individual clinic.

About us

SAQR MED is part of the SAQR AI product line

SAQR AI is a technology group building applied artificial intelligence for critical industries — energy, water, industry and medicine. We combine deep domain expertise with modern AI methods and take ownership of the customer's operational outcome.

One platform, multiple verticals

SAQR EDGE — autonomous perception for critical infrastructure (solar, desalination, oil & gas, construction).

SAQR MED — computer vision for early visual diagnostics (liver, heart, pancreas, kidneys).

Each line has its own team, its own delivery model and its own regulatory path. A single brand and a shared engineering foundation.

Legal entity

All commercial relationships are conducted through an entity registered in the Kyrgyz Republic.

Company	SAQR AI LLC
Registration number	329540-3301-OOO
Tax ID (INN)	00106202610099
OKPO	34970908
Registered address	141/1 Patrice Lumumba St., Bishkek, Kyrgyz Republic
Registration date	1 June 2026, Chui-Bishkek Department of Justice
Director	Evgeny A. Korshunov

Go to saqr-ai.com →

A trained model. Transferred to the customer. Running locally.