Organization isolation
Users cannot access records that belong to another organization.
Research and demonstration prototype
From a CT scan to a clearer clinical decision, in one workflow.
SANAD helps radiologists review potential pulmonary nodules, correct their location and measurements, and follow each case without moving between disconnected systems.
Clinical decision-support prototype. The final decision belongs to the radiologist.
9.4 mm
Model confidence95%Supporting information only
84independent test scans
95.1%test sensitivity
74automated tests passing
5clinical workflow steps
The problem
An AI result alone is not enough.
A radiologist needs to inspect the finding, adjust its measurement, document a decision, and know what happens next. SANAD brings those tasks into one clear experience.
BeforeScattered tools and files
Manual handoffs, unclear steps, and model output separated from the clinical decision.
With SANADOne clinical workflow
One required action at each stage, while the final decision remains with the clinician.
The workflow
Five steps, without confusing menus.
The system presents the action required now and moves the case to the next stage.
- 01Patient
Select an existing record or register a new case.
- 02Imaging
Upload a CT study and validate its files safely.
- 03Analysis
Detect potential findings with a MONAI model.
- 04Clinical review
Accept, correct, or reject each candidate.
- 05Follow-up
Create a plan and notifications automatically.
Interactive demo
See what the radiologist sees at every step.
This demonstration uses synthetic data and does not represent a real patient.
Step one
Start with the patient
SANAD shows only the case information needed for the task and restricts access by organization and role.
- Synthetic case identifier
- Clear and concise record
- Role- and organization-based access
Test results
High sensitivity, with clear room for improvement.
The model was tested on 84 held-out scans containing 102 nodules. It detected 97 and missed 5, while producing false positives that require radiologist review.
How should this result be read?
SANAD demonstrates that the technical workflow operates end to end. These numbers do not represent clinical validation or approval for clinical use.
Sensitivity95.1%97 of 102 nodules
Positive predictive value31.5%
False positives per scan2.51
The decision threshold was selected on validation data before it was applied to the test set.
Safety and responsibility
Designed around clinical review and case privacy.
Protected medical files
DICOM studies are stored outside the public media directory.
Required human review
Every result remains a candidate until a clinician accepts or corrects it.
Audit trail
The system records review decisions and important changes.
Open-source project
Open the project on GitHub
SANAD understands the workflow and leaves the decision to the clinician.
Explore the source code, model card, and security tests in the repository.