AI expert systems for radiology · Prague

aEyes

by MontesMedica

AI expert systems for MRI. Each module is a specialised second reader for one body region. The first module reads knee MRI: abnormal finding, anterior cruciate ligament tear and meniscal tear. Our focus is sports injuries in children and adolescents.

Knee MRI · now Spine MRI · 2027 Histopathology · 2028
Second reader
Supports the radiologist's decision, never replaces it. The physician always makes the final call.
In the hospital
Runs on a single GPU workstation inside the hospital. Patient data never leaves the site.
Reasons shown
For every finding, it shows the slices that led to it.
EU MDR
Designed from day one as a Class IIa medical device.

First module · aEyes-knee

Knee MRI: more scans every year,
and reporting can't keep up

Volume

Three planes, dozens of slices in every study

Knee MRI is among the most frequent MRI examinations. Scanner numbers keep growing; radiologist numbers do not grow with them.

Decisions

Two findings decide what happens next

An ACL tear and a meniscal tear decide between conservative treatment and surgery. In the Stanford MRNet reader study, model assistance raised clinicians' specificity for ACL tears by 4.8 percentage points: fewer false positives in patients without a tear.

Source: Bien et al., PLOS Medicine 2018 (MRNet reader study)

How it works

From images
to a second read

01

Knee MRI study

A standard examination in three planes (sagittal, coronal, axial), as DICOM straight from PACS.

02

Every slice

A neural network processes each slice together with its neighbouring slices.

03

The whole series

The model learns which slices matter and weights them accordingly. Those are the slices it later shows the radiologist.

04

Three planes together

Results from all planes are combined into three probabilities: abnormal finding, ACL tear, meniscal tear.

05

Second read

Probabilities, highlighted slices and a structured report go back into the tools the radiologist already reports in.

The aEyes suite

One expert system
per organ

Every module stands on one platform: connection to PACS and the DICOM viewer, computation on the workstation, operational monitoring and an MDR quality system. Each new module reuses everything already built.

Now · knee MRI

aEyes-knee

Abnormal finding, ACL tear, meniscal tear. The first module and the regulatory template for all that follow.

2027 · spine MRI

aEyes-spine

Lumbar spine. Same platform, same hospital installation, new module.

2028 · histopathology

aEyes-path

Digitised slides at gigapixel resolution. Same principle: the model learns which parts of the image decide.

Shared platform

One platform

From model training to the reading room. It plugs easily into the tools a department already uses: PACS, RIS, DICOM viewer. One GPU workstation per site, no patient data in the cloud.

Roadmap

Demo, pilot, CE mark

On public data we show that the model reaches published results. The certified device is then trained on prospectively labelled studies from partner hospitals.

Q4 2026

Proof of concept

  • Accuracy on par with published results on the MRNet dataset
  • External validation on independent data
  • Demo running on a single GPU
H1 2027

Clinical pilot

  • Installation at the pilot site
  • Reader study at a partner hospital
  • Data agreements
H2 2027

MDR Class IIa

  • ISO 13485 quality management system
  • Technical documentation and clinical evaluation
  • Spine module proof of concept
2028

CE mark

  • CE marking for aEyes-knee
  • First deployments in Czechia
  • Histopathology module proof of concept

Partnership

We are looking for partner hospitals, radiologists and investors

If you are interested in a pilot deployment, a reader study or a data partnership, get in touch.

aEyes is under development and is not yet a certified medical device. It is not intended for clinical use.

info@montesmedica.cz