Lävi Lävi Clinical Suite · Longitudinal infrastructure for brain MRI
Make interval change easier to see.
Lävi Delta is our first product: a second reader for follow-up brain MRI. It gathers the candidate changes between baseline and follow-up into one auditable review — and when two studies cannot be compared on technical grounds, it says so and withholds the result.
Clinical validation and regulatory development in progress · not CE-marked
-
Implemented
Working end-to-end system
The longitudinal pipeline, DICOM export and dedicated reader workstation are implemented and tested — and verified to work together in a full production-form deployment.
-
Deployment model
Built for hospital deployment
On-premise architecture with standard DICOM in and out, returning results into the existing PACS workflow. In that deployment, no image data needs to leave the institution.
-
In preparation
Clinical validation next
An independent multi-reader study is in preparation — and the reading infrastructure it needs already exists. The plan.
01 — The problem
Why longitudinal review is hard
Reading a follow-up scan means answering a comparative question: has anything changed since last time? In practice that is harder than it sounds. The two examinations were often acquired months apart, on different scanners, at different slice angles, under different protocols. The reader scrolls two stacks side by side and holds the difference in memory — case after case, on clinical time.
Slow, diffuse change is the kind most easily missed. A finding that grows a little between every visit can look stable at each single comparison. A subtle new finding can sit at the edge of a slice that has no true counterpart in the prior study.
Underneath both sits a quieter problem: sometimes the two studies simply cannot be compared, because positioning or acquisition differs too much. Software built for follow-up reading has to recognise that situation. Papering over it would be worse than offering no help at all.
And this work compounds. Surveillance protocols add examinations to each patient's history year after year, while the review itself is still a side-by-side human comparison. Repeated imaging is a longitudinal data problem, not a stack of single-scan reads.
Fig. 2 — The geometry problemschematic; not patient data
02 — The workflow
What Lävi Delta does
-
Longitudinal pair received
Baseline and follow-up arrive over standard DICOM, the same way any study moves through the department. Processing starts on its own, with no operator interaction, and in the intended on-premise deployment the images stay inside the hospital network.
-
Registration and quality assessment
The follow-up is registered to the baseline. Before any analysis, the pair itself is examined: are these two studies technically fit to compare? The answer is graded — report, report with caution, or a refusal to report the comparison at all.
-
Candidate change analysis
Candidate regions of change are identified and ranked, each with its location, volume and direction. New or growing change and resolving change sit in separate lanes, because they mean different things to the reader.
-
Radiologist review
The reading radiologist goes through the candidates and accepts or rejects each one. Lävi Delta makes no diagnostic decision; the radiologist still reads the whole study, with the candidate list alongside.
-
Structured, auditable record
What returns to PACS is ordinary DICOM: segmentation objects, overlay series and a structured report, each bound to the provenance of the run that produced it.
03 — The platform
Built around the comparison, not the model
Most imaging algorithms solve one narrow, single-scan analytical task. The clinically hard part of longitudinal reading sits around any such model: deciding whether two examinations can validly be compared, preserving provenance, managing uncertainty, integrating with PACS, presenting change for a human decision — and withholding the answer when the evidence does not support one.
That surrounding layer is what Lävi is building, and it is deliberately not tied to one indication: ingestion of longitudinal studies, registration and comparison, quality control with explicit abstention, candidate-change analysis, provenance, standard DICOM and PACS integration, the auditable radiologist review environment, the reader-study infrastructure, and the validation and regulatory architecture around all of it. Lävi Delta is the first product built on that layer — and the first that will carry it through independent clinical validation.
Who it is for
Lävi Delta is being built for radiology departments and imaging organisations that read longitudinal brain MRI routinely. It is designed to fit the workflow a hospital already runs: studies arrive over DICOM, analysis stays inside the hospital network, and review returns to the radiologist through familiar imaging infrastructure. Deployment and early collaboration →
04 — Clinical domains
Where interval review matters first
-
Stratum I
Multiple sclerosis
Surveillance MRI at regular intervals, where new activity between two visits can change treatment.
-
Stratum II
Primary brain tumours
Follow-up after treatment, where slow growth has to be told apart from a stable post-treatment appearance — across a long series of examinations.
-
Stratum III
Brain metastases
Follow-up after treatment, where a small new finding and change in a known one carry equal weight.
Lävi Delta is being developed for adults (18+) undergoing longitudinal brain MRI surveillance. These three settings are the strata of the planned reader study. It is not intended for paediatric imaging, non-brain anatomy, other modalities, or acute stroke triage.
05 — Validation
Clinical validation programme
Does a radiologist read follow-up brain MRI better with Lävi Delta than without it? That is the question, and we intend to answer it with an independent reader study rather than a marketing sentence.
We would rather prove it than assert it.
In the planned study, one hundred retrospective cases from the three settings above will each be read by several radiologists — first without the software, later with it. Reading every case both ways lets the study measure the software's effect on the reader while accounting for how much readers and cases differ. The reference standard and the statistical analysis are planned to be independently defined and reviewed, and the final design is subject to methodology review.
Clinical validation programme in preparation
06 — Safety philosophy
Quality and abstention
-
When the comparison cannot be trusted, there is no result
If the prerequisites for a valid comparison are not met, Lävi Delta withholds the longitudinal result and reports the refusal itself. An honest "cannot compare" is worth more than a number nobody should rely on. This behaviour is implemented and tested today, not planned.
-
Registration success is not proof of correspondence
Equal geometry, a converged registration, a familiar method name — all of it is recorded as provenance, none of it is treated as evidence that the anatomy actually corresponds. Where correspondence is unsupported, that lane abstains.
-
Every result carries its provenance
Method, policy identity, input hashes and environment are recorded on every run and bound to the exported objects — so any result can be traced back to exactly the run that produced it. How far each processing path can be replayed is spelled out on the technology page.
-
A second reader, not a substitute
An empty candidate list is not evidence that nothing changed. The radiologist reads every study independently, exactly as before; the software's job is to support that reading, never to stand in for it.
The comparison meets its prerequisites and is reported.
The comparison is reported together with its identified limitations.
The longitudinal result is explicitly refused, and the refusal is stated.
07 — Technology
Built to be checked
-
Runs inside the hospital
Deployed on-premise, speaking standard DICOM in both directions: studies in, segmentation objects and reports out, straight into the existing PACS workflow. The images stay in the building.
-
Traceability is machine-checked
If a requirement cites a test that does not exist, or a controlled document no longer matches the code, the build fails. We added those guards after our own audits caught traceability rows that did not hold — once was enough.
-
Verified end to end
One chain, exercised in a full production-form deployment: pipeline run, exported package, image store, reader workstation. Rejected cases were verified to stay rejected rather than slipping through.
08 — Company
Lävi Clinical Suite
Lävi Clinical Suite OÜ is a medical software company in Tartu, Estonia. We are building Lävi Delta as a medical device under the EU Medical Device Regulation — slowly enough to do it properly.
-
Andreas Müürsepp — Founder & CEO
Neuroradiologist. Andreas reads longitudinal brain MRI clinically and drove the design of the Lävi Delta pipeline from its first version, so the reading-room problem and the software stay unusually close.
-
Alvar Haug — CTO
Runs infrastructure, deployment and security — the plumbing a clinical system stands on, and the first thing a hospital IT department asks about.
The company is deliberately built on three legs: neuroradiology and clinical workflow, software and infrastructure engineering, and independent clinical validation — the reference standard and the statistical analysis of the planned reader study are planned to be independently defined and reviewed, not decided by us.
Company development is supported by participation in the Health Founders Estonia DEVELOP programme.
The team is growing. If your work touches medical imaging, clinical software or regulatory affairs, write to us.
09 — Roadmap
What comes next
Lävi Delta is the first product, not the whole plan. What it establishes — registration, quality control and abstention, provenance, DICOM integration, the review environment, the validation methodology — is the foundation each subsequent longitudinal product reuses rather than rebuilds.
Lävi Delta
Longitudinal brain MRI review. The present development, clinical-validation and regulatory focus — and the platform everything after it stands on.
Lävi MS
A dedicated multiple sclerosis imaging workflow, building on the longitudinal, quality-control and validation infrastructure established with Lävi Delta. Its own programme begins once Delta's reaches the required maturity.
Further longitudinal workflows
Subsequent longitudinal brain MRI products can reuse the same validated quality-control, provenance and deployment infrastructure. None are announced beyond Lävi MS — the platform is the point, not a product list.
10 — Development status
What exists, and what is not yet proven
Built and verified
- The on-premise pipeline: standard DICOM in, automated processing, standard DICOM objects out.
- Quality-control gating with graded outcomes, including explicit abstention.
- Per-run provenance, bound to the exported results.
- A dedicated reader workstation for the planned study — with arm blinding, washout enforcement, role-based access and an audit trail, implemented and tested.
- Machine-checked traceability from requirements to tests, enforced in the build, with a large continuously run automated test suite.
- The full producer-to-reader chain, verified in a full production-form deployment.
Not yet proven
- Clinical performance. Whether Lävi Delta changes reader performance has not been established — that is what the planned reader study exists to measure.
- Cross-vendor and cross-protocol comparison: a design goal and an explicit requirement of the validation programme, not an established capability.
- Capabilities still under technical evaluation, which are not part of any claim.
- Regulatory conformity: the conformity-assessment stage lies ahead, as the statement below spells out.
Lävi Delta is under development. It is not CE-marked, is not available for sale, and must not be used for clinical decision-making. Information on this site describes a product in development and its planned evaluation, not the performance of a marketed medical device.
Status as of 2026-08-21
11 — Contact
Talk to us
One address, four doors — the subject line tells us which one you are knocking on. The founder answers his own email.
-
Hospitals & clinical collaboration
The validation programme, early-site collaboration, or follow-up imaging in general. Write about clinical collaboration →
-
Careers
Medical imaging, clinical software, regulatory affairs. Ask about working here →
-
Investment
The evidence plan and the technical architecture, examined in detail. We maintain dated internal development, validation and evidence-planning documentation; selected material can be shared in qualified conversations under NDA or controlled data-room access. Start an investor conversation →
-
Lävi MS
The planned MS programme, and how the follow-up read looks from where you sit. Write about MS imaging →