Home/Security

Patient data,handled like it matters.

OncoBoard holds the clinical detail a tumour board needs and nothing that names the patient. This page sets out what is stored, who can change it, where it travels and what your organisation keeps control of.

0

Fields for a patient's full name

scrypt

Password hashing with a per-user salt

12 h

Before a session expires

11

Clinical actions written to the audit log

01 · Data

De-identified by design.

A case is identified by its MRN and the patient's initials. The case record has no field for a full name, date of birth or address, so there is nowhere to type one.

  • Stored: MRN, initials, age, sex, board, urgency and status.
  • Stored: TNM, AJCC stage group, ECOG, pathology, IHC, NGS variants, labs, treatment history and trial matches.
  • Stored: saved measurements, votes, minutes, case chat and the audit log.
  • Not sent to the server: uploaded DICOM files. They are parsed in the clinician's browser.
Case record · as stored Example
MRN
ONC-2026-4102
Patient
R.K. · 64
Board
Thoracic
Urgency
Priority
Staging
T2b N2 M0 · IIIA
ECOG
1
NGS · Tier I-A
EGFR p.Leu858Arg
Status
In Review
Full name · date of birth · address
The case record has no field for these
Not stored

DICOM headers can carry a name. Uploaded series stay in the browser and only the measurements you save reach the server. The viewer still shows header tags, so upload de-identified exports from PACS.

02 · Access

Every clinician signs in as themselves.

Accounts carry the clinician's specialty, title, institution and whether they chair the board. Votes, sign-offs and measurements are tied to that account, so the record shows who said what.

Password storage
scrypt · 16-byte random salt per user · 64-byte key
Password check
Constant-time comparison
Session token
256-bit random bearer token
Session lifetime
12 hours · revoked at sign-out
Signed-in only
Every write and every AI request
Roles
Six specialties · chair flag
Signed-in session Example
Diagnostic Radiology
Consultant · Thoracic Board
Signed in
Chair
No
Token
9f3c…a71e · 64 hex
Expires
in 11 h 42 m
Can vote
Yes
Vote · ONC-2026-4102
Recorded under this account and specialty
Audited
03 · Audit

Who did what, and when.

Each clinical change writes an audit entry with the clinician, their specialty, the action, the record it touched and the time. Your IT team reads it in the Audit & Database view.

Case createdCase updatedMeasurement savedMeasurement deletedSession scheduledVote castConsensus signedAI synthesisAI slice readAI minutes draftedClinician registered
Audit log · ONC-2026-4102 Example entries
07:46:12
Measurement saved
Diagnostic Radiology · Target lesion 1 · 24.5 mm
Imaging
07:49:30
AI slice read
Diagnostic Radiology · slice 42
AI
07:52:40
Vote cast
Surgical Oncology · not for upfront resection
Vote
07:54:18
AI minutes drafted
Medical Oncology · chair
AI
07:55:03
Consensus signed
Medical Oncology · chair · 5 of 5 votes
Ratified

AI governance

The AI can draft.
It cannot decide.

OncoBoard's four assistants use the Google Gemini API. These are the limits they work inside.

Called on request

The server calls the model only when a clinician presses an assistant button. Nothing runs in the background, and without an API key configured nothing is sent at all.

What leaves the server

Case synthesis and case questions send the de-identified case context. A slice read sends a rendered image of the current slice, not the DICOM file. Minutes send the votes and recent case discussion.

Labelled and disclaimed

Every output is headed "AI Decision Support" and ends with a disclaimer, wherever it appears.

Cannot act on the record

Assistants cannot vote, sign off or edit a case. A clinician reads the draft and does each of those.

On the audit log

Synthesis, slice reads and drafted minutes are written to the audit log with the clinician who asked.

Your key, your choice

AI runs on an API key your organisation configures on the server. Leave it unset and the rest of OncoBoard works as normal.

Architecture

One server.
One database.

OncoBoard is a React 19 web app, a Node.js and Express server with a WebSocket hub, and a MySQL database. The recommended deployment serves it over HTTPS and WSS.

OncoBoard data flow Clinician browsers connect over HTTPS and WSS to a Node.js and Express server with a WebSocket hub. The server reads and writes a MySQL database. Both sit inside the deployment your organisation approves. The server calls the Google Gemini API only when a clinician requests an assistant, sending de-identified case context. Deployment your organisation approves Clinician browser React 19 web app DICOM parsed locally Signed-in sessions HTTPS WSS Node.js + Express REST API · session check WebSocket hub · presence, sync Audit log writer scrypt · 12 h sessions MySQL Cases, votes, measurements, audit log Gemini API AI assistants On request only De-identified context
Clinical data in transitAI request, only when a clinician asksInside your approved deployment

Shared responsibilities

What we provide.
What stays with you.

Security is split between the software and the organisation that runs it. This is where the line sits.

AreaOncoBoard providesYour organisation provides
HostingA Node.js app with a MySQL schema that runs on standard hostingThe server, database and network it runs on, or approval of the host
Encryption in transitAn app and WebSocket service that run behind HTTPS and WSSThe TLS certificate and the HTTPS endpoint
BackupsAll case data in one MySQL databaseBackup schedule, retention and tested restores
User accountsClinician accounts with specialty, title, institution and chair flagWho gets an account, and removing leavers
SessionsRandom bearer tokens that expire after 12 hoursDevice policy for shared reading-room workstations
De-identificationNo field for a full name, date of birth or addressDe-identified DICOM exports and identifier-free free text
AIOn-request calls, labelled output, no write access to the recordThe decision to use AI and the API key under your own agreement
AuditAn audit entry for every clinical change and AI draftWho reviews the log, and how often
GovernanceA walkthrough of this page with your teamDPIA, clinical safety sign-off and local operating procedures

Information governance

Bring your questionnaire.

Send your security or DPIA questionnaire with your demo request. We will go through it with your information governance lead in the session.

Book a Demo

Questions

Asked by
governance teams.

Where is patient data stored?
In the MySQL database of the deployment you approve. Uploaded DICOM series are parsed in the clinician's browser and are not sent to the server; only measurements you choose to save are stored.
Does OncoBoard store patient names?
The case record has no field for a full name, date of birth or address. Cases carry an MRN and initials. DICOM headers and free-text fields are the two places a name could still appear, which is why we ask for de-identified exports and a local rule on free text.
Is any data sent to an AI provider?
Only when a clinician presses an assistant button. The server then sends the de-identified case context, or a rendered image of the current slice, to the Google Gemini API using the key your organisation configures. With no key configured, no AI request is made.
Can clinicians type identifiers into free text?
Yes. The clinical summary and case chat store what is typed. Your local policy should say identifiers stay out of free text, in the same way it does for email and meeting notes today.
What does the audit log record, and what doesn't it?
It records clinical changes and AI drafts: cases created and updated, measurements saved and deleted, sessions scheduled, votes cast, consensus signed, AI synthesis, slice reads and minutes, and new clinician accounts. It does not record who viewed a case.
Which certifications does OncoBoard hold?
This page describes only the controls that exist in the product today. We cover certification and assurance status in the demo, with your information governance lead in the room.

Walk your IG lead through it.

Thirty minutes with this page, your questionnaire and a de-identified sample case.

Book a Demo