Product
AIVRA's medical AI platform for chest radiology. One analysis module today, a full suite over time.
The modules
Click a module to see the detail: its scope, what it produces, and what it does not do.
Chest detection
Twenty classes across eight physiopathological families: cardiovascular, air, fluid, mass, parenchyma, infectious, skeletal, and a reference class for images with no detected anomaly. Every detection is localised on the image and carries a confidence score. The reference class states what the model did not find, not what does not exist: an absence of detection is never an absence of pathology.
Automated report
The report lays the detections out in a fixed structure — exam, technique, findings — so that two reports can be compared. The radiologist corrects it by keyboard or dictates it aloud: the transcription is formatted into the same structure, with no re-entry. The text produced remains a draft until it is reviewed.
3D reconstruction
Volumetric reconstruction from a series of images, for a spatial reading the frontal projection cannot give. This module is on the roadmap: it is neither built nor dated. It is named to say where the platform is heading, not to be ordered today.
How it works
The path of a single exam, from the moment the image is taken to the report the radiologist signs off.
Step 1 of 4
Acquisition
The image reaches OXALIA at the very moment it reaches the PACS: nothing extra is asked of the radiographer, and no workstation has to be installed in the room. Images arrive in standard DICOM format, the one your installation already produces. The mobile-unit scenario is designed to work over an unstable or intermittent connection.
Step 2 of 4
OXALIA analysis
The analysis covers the frontal chest radiograph. The model was developed in Tunisia on Tunisian data, which matters more than it may seem: disease prevalence and imaging habits differ from one country to the next. Every detected anomaly is localised on the image rather than merely named — that is what lets the radiologist check it, and disagree.
Step 3 of 4
Prioritisation
Exams are not returned in the order they arrive, but in the order that matters: an image carrying suspicious signs moves ahead of routine cases. This is the only place where OXALIA acts on your organisation. No exam is removed from the list and none is set aside — only the reading order changes.
Step 4 of 4
Medical validation
The radiologist opens the preliminary report, corrects it, completes it or rewrites it. They can also dictate it: OXALIA transcribes the voice and formats the text. Nothing leaves OXALIA without this step — it is both the limit the solution sets itself and what its regulatory status requires.
Diagnostic aid — the final decision rests with the radiologist.
At a glance
- Modality
- Frontal chest radiograph, in standard DICOM format.
- Coverage
- 20 classes across 8 physiopathological families.
- Analysis time
- Under two minutes per exam.
- Integration
- Existing PACS and DICOM systems, with no infrastructure change.
- What it produces
- A structured preliminary report, editable by keyboard or by voice.
- Validation
- Mandatory. No report leaves OXALIA without the radiologist's review.
The exact conditions behind the timing figure, and the technical prerequisites for integration, are established during a preliminary technical discussion.
The same detection module is not deployed the same way in an emergency department, in teleradiology, or in a mobile screening unit.
See OXALIA on your own workflow.
See OXALIA in a setting close to your own radiology workflow.