Skip to content
healthcare ainational healthcare organization

Digitization and analysis of medical reports.

600,000 sensitive documents digitized, anonymized and enriched with tags and KPIs. Interactive dashboards and an AI Assistant for natural-language queries.

The project in numbers

600,000
documents digitized and anonymized
200,000
documents a year, previously processed by hand
3
outcome classes: invalid · pediatric · pathological
the system, in cross-sectiona single flow
quality gatefour steps per lead

How you decide whether a trace is usable.

No hand-tuned thresholds: the reference is the set of reports the client has already validated. A trace enters the flow only if its profile falls within the percentiles of the good ones.

01
Lead extracted

Each lead cropped from the PDF, at its own scale.

02
Binarised

Grid and background gone: the trace remains, in black and white.

03
Angles and derivatives

Point by point: slope and normal to the trace.

04
inside the percentiles · p10–p90outside: to review
Classified

The profile compared against the percentiles of valid reports.

noteExample trace, reconstructed to illustrate the method: no real clinical data appears on this site.
what changes

Data extraction, validation and AI analysis in one continuous flow: less manual work for staff, more time on the cases that matter.

Patients who get answers faster. Doctors who spend more time on care.