2.0 Mediscan IA

Admission, in 3 seconds.

Scan the prescription. Mediscan AI pulls out the patient, the exam, the allergies and the Tardoc codes — and offers the slot that fits the schedule.

mediscan://extract · 200 OK · 412ms
LIVE
Source · Prescription
Cabinet du LacDr. M. Meier
14 NOV 2026N° 28842
Cabinet
du Lac
★ ★
Extraction · structured
Latency 412ms
PatientRochat, Paul · 1978✓
PrescriberDr. M. Meier✓
ExamBrain MRI✓
ContrasteGadolinium✓
UrgencyModerate · 7d✓
Tardoc39.0020 · 39.0040✓

Recognition, no strings attached.

Handwritten, printed, low-quality photo, PDF scan — Mediscan AI structures any prescription in under a second.

2.1 Extraction →
→
Pat.Rochat, P.
Exa.IRM C+
Urg.Moderate
Tar.39.0020
FIG. 2.1

OCR on medical handwriting

Our engine was trained on 2.4M prescriptions (FR · DE · IT). It handles fast handwriting, clinical abbreviations and crossed-out lines with no manual rework.

p50 latency
412ms
0ms1s
FIG. 2.2

Pipeline under 500 ms

OCR rebuilt in v4, distilled structuring model, Tardoc checks built in. Handwritten prescriptions average under 500 ms, vs 1.4 s in v3.

✎
Handwritten
A
Printed
📷
Photo
PDF
PDF
DCM
DICOM SR
FAX
Fax
FIG. 2.3

Every format, one pipeline

Smartphone photo, front-desk scan, referring PDF, hospital fax — Mediscan ingests and structures all the same. Radiologists have one place to look.

PatientRochat, P. · 1978✓
PrescriberDr. Meier✓
ExamIRM C+✓
AllergiesIode, latex✓
Urgency7 days✓
Tardoc39.0020✓
FIG. 2.4

Normalized clinical fields

Patient, prescriber, exam, contrast, allergies, urgency, codes: every critical field comes out normalized as FHIR R4, ready to flow into the RIS.

Checks. Cross-references. Alerts.

Mediscan doesn't just extract — it cross-checks every field against the patient record, known contraindications and billing conventions before suggesting a slot.

2.2 Clinical safety →
!
Contraindication detected

Gadolinium allergy on file

The patient declared a moderate reaction in March 2024. Contrast MRI is not recommended — propose a non-contrast MRI or another protocol.

FIG. 2.5

Allergies and contraindications

Auto cross-check against the patient record, Swissmedic drug list and exam history. Contraindications surface before the appointment is confirmed.

Mediscan RIS DPI Pharma Tardoc FHIR Swissmedic
FIG. 2.6

Cross-check with patient record

Mediscan plugs into the RIS, hospital EMR, drug list and Tardoc/FHIR referentials. No manual entry, no duplication, nothing missed.

Prescribed examIRM cérébrale + C
→
39.0020 · 39.5510
Senior readDr. Favre
→
39.0010
Contrast agentGadolinium 15ml
→
39.5650
FIG. 2.7

Auto Tardoc mapping

Every prescribed exam, every product, every act comes out with the right Tardoc positions — checked against tariff conventions before billing.

97%
Extraction confidence
FIG. 2.8

Per-field confidence score

Every extracted field carries its confidence score. Below the threshold set by your practice, Mediscan asks for a human check — no blind entry.

The slot that fits the patient.

The AI doesn't pick the first available slot — it weighs urgency, modality, room load, home distance and referrer preferences to suggest the best one.

2.3 Slot scoring →
Mar 09:15Room 2 · Lausanne82
Mar 11:30Room 1 · Lausanne · best fit96
Mer 08:45Room 3 · Geneva74
Mer 14:00Room 2 · Lausanne71
Jeu 10:00Room 1 · Geneva65
FIG. 2.9

Top 5 slots, scored

Every open slot is scored in real time based on prescribed urgency, room load, available modality and patient history. The front desk validates in one click.

Urgency
92
Room load
88
Distance
74
Referrer
61
Preference
58
FIG. 2.10

Per-practice weighting

Scoring weights are configurable — an on-call practice prioritizes urgency, an ambulatory practice prioritizes room load. The model fits your operation.

Your data never leaves your infrastructure.

The model improves with every practice, but the prescriptions stay with you. Federated learning, HDS hosting, end-to-end encryption — by default.

2.4 Sovereignty →
DC 1 DC 2 DC 3
FIG. 2.11

HDS hosting · triple replication

Encrypted triple replication across redundant datacenters, 0 outsourcing to a US public cloud. Yearly audit by EY, report available under NDA.

Mediscan model Cabinet A Cabinet B Cabinet C Cabinet D CHU A CHU B
FIG. 2.12

Federated learning, zero exposure

Each practice trains locally, only encrypted gradients are pushed. The global model improves with every prescription — without any patient data ever leaving your infrastructure.

Extraction accuracy97%
p50 latency412ms
Saved per admission12min
Patient data exposed0%

One admission, three seconds.
Try it now.

Try Mediscan Talk to the team