What AR Surgical Navigator does
AR Surgical Navigator helps surgical teams rehearse, brief and document a case with a patient-specific 3D organ model and AI assistance — without ever exposing protected health information to the AI.
Hospitals upload a segmented STL organ mesh (and optional DICOM metadata) per procedure. The model renders in a 3D viewer that anyone on the team can orbit and inspect, and on supported devices it can be overlaid on the patient in AR against a printed marker for orientation. During a case, a surgeon can ask anatomy questions hands-free by voice and hear the answer read back; run a WHO-style safety checklist with AI guidance on each step; and, afterwards, dictate a de-identified operative narrative that the AI turns into a structured note. Residents can generate practice quizzes and explore the same 3D anatomy.
Audience
Surgical teams
Consultants and operating surgeons who want a shared 3D reference, hands-free anatomy answers and a structured safety checklist for a case.
Surgical residents
Trainees who explore procedure anatomy in 3D and test themselves with AI-generated quizzes at chosen difficulty.
Teaching hospitals
Departments that want a compliant, de-identified tool for pre-op briefing, intra-op reference and post-op documentation across many procedures.
End-to-end flow
Upload imaging
A segmented STL organ mesh (and optional DICOM for non-PHI metadata) is uploaded per procedure. DICOM is parsed client-side and PHI tags are stripped.
3D organ model
Three.js STLLoader renders the mesh in an orbit-controlled viewer. If no STL exists, a labelled representative placeholder organ is shown.
WebXR overlay
On supported devices the model can be placed in AR via hit-test, using the printed marker as a visual registration guide over the patient.
Voice anatomy AI
The surgeon asks a question hands-free; the AI answers using only de-identified procedure/anatomy context and speaks the answer back.
Safety checklist
A WHO-style checklist is ticked through, with optional AI guidance and safety checks on each step (procedure and step titles only).
Operative note
The surgeon dictates a de-identified narrative; the AI drafts a structured note (findings, summary, post-op instructions) to print or save.
Resident training
Residents generate MCQ quizzes per procedure and difficulty, reveal answers with explanations, and score themselves.
Renew annually
Each Razorpay payment keeps every module live for one year for the whole hospital.
Every page and what it does
| Page | Nav | What it does |
|---|---|---|
| SRN_SurgeonDashboard.aspx | dashboard | Post-login home. Hospital header (name/tier from SRN_usp_GetHospital), stat cards from SRN_usp_GetDashboard (patients, procedures, uploads, sessions, surgeons), and quick CTAs into patient prep, AR navigation and operative notes. |
| SRN_PatientPrep.aspx | patients | Manage de-identified patients (code, age, sex, hospital-internal display name marked “not sent to AI”) and per-patient DICOM/STL uploads. DICOM is parsed client-side to show modality/dimensions; only PHI-stripped metadata is stored. Assign a procedure to prep an intra-op session. |
| SRN_ProcedureSetup.aspx | procedures | Manage procedures (name, body region, description, template flag), upload the organ STL model and marker image, and edit each procedure’s ordered safety-checklist steps (step number, title, instruction, safety note). |
| SRN_ARNavigation.aspx | arnav | The hero page. Loads the procedure’s STL model in a Three.js scene (orbit + optional WebXR overlay), with hands-free voice anatomy Q&A and an intra-op session (start/end). A persistent de-identified-mode reminder is always visible. |
| SRN_CheckList.aspx | checklist | Pick a procedure and work through its steps as an interactive WHO-style surgical safety checklist. Each step has an optional AI-guidance button that returns guidance, safety checks and a next-step hint (procedure and step titles only). |
| SRN_PostOpReport.aspx | postop | Pick a session or a procedure + de-identified patient (code/age/sex only), dictate or type the operative narrative, and generate a structured operative note (note, findings, summary, post-op instructions) in a print-friendly layout. |
| SRN_ResidentTraining.aspx | training | Pick a procedure and difficulty and generate an interactive MCQ quiz (options, reveal answer + explanation, score). Residents can also open the procedure’s 3D model to explore anatomy. |
| SRN_Subscription.aspx | subscription | Shows the hospital’s current plan, validity and days remaining, and runs Razorpay checkout for the Rs.50,000 / hospital / year annual plan. Requires the settings capability. |
SRN_DicomUploadHandler.ashx and SRN_ModelUploadHandler.ashx (session-gated uploads), and SRN_RazorpayWebhook.ashx (HMAC-verified payment webhook → SRN_usp_SaveSubscription).What the AI does — and what it never sees
All AI calls go through SRN_ClaudeHelper, which routes to the configured provider
(Groq by default, Claude/Anthropic when configured), retries for valid JSON, and runs a
ScrubPHI defence-in-depth pass on free-text inputs. Every method receives only
de-identified data.
AnatomyQuery
Hands-free voice anatomy answers on the AR Navigation page. Inputs: question, procedure name, anatomy region. Returns an answer, key structures and a caution.
ChecklistGuidance
Per-step guidance on the checklist. Inputs: procedure name and current step title only. Returns guidance, safety checks and a next-step hint.
OperativeNoteFromDictation
Drafts a structured operative note. Inputs: procedure name, de-identified dictation, patient age and sex. Never receives a name or identifier.
ResidentTrainingQuiz
Generates MCQ practice questions. Inputs: procedure name and difficulty. Returns questions, options, correct answers and explanations.
De-identified by design
The #1 rule: no protected health information ever reaches the AI. The database stores patient records for the hospital’s internal use, but every value passed to an AI method is de-identified. AI-facing code uses only a patient code (for example PT-0007), age in years, sex, procedure name, anatomy region and non-identifying clinical text.
No identifiers to AI
SRN_Patients.DisplayName and .Notes are hospital-internal only and are never included in any AI prompt. Age (under 90) and sex are permitted under HIPAA Safe Harbor.
Client-side DICOM, PHI stripped
DICOM files are parsed in the browser with dicomParser to read only non-PHI metadata (modality, rows/cols, study description). Patient-identifying DICOM tags are not displayed or stored.
STL meshes, not volumetric reconstruction
3D models are pre-segmented STL meshes supplied by the hospital. Full volumetric DICOM-to-mesh reconstruction in the browser is out of scope and documented as a future step.
Defence in depth
SRN_ClaudeHelper.ScrubPHI runs over free-text inputs (voice transcripts, dictation) as a safety net — but pages still only pass de-identified data in the first place.
Persistent UI reminder: the AR navigation and operative-note pages show a standing note — “De-identified mode: do not speak patient names or IDs — this data is processed by AI.”
Important safety disclaimer
AR Surgical Navigator is a visualisation, education and training aid. It is not a certified, cleared or regulated surgical navigation, guidance or diagnostic device. Its 3D models, AR overlays and AI outputs are for orientation, briefing, documentation support and learning only, and must not be used as the basis for clinical decisions or intra-operative registration. WebXR/AR placement uses hit-test and a printed marker as a visual guide — it is not sub-millimetre surgical registration. Clinical judgement, approved imaging and certified equipment remain the source of truth.
Architecture
Backend
ASP.NET Web Forms + VB.NET on .NET Framework 4.7.2. AJAX-only PageMethods returning JSON; no full-page postbacks for data.
Database
SQL Server 2014 (shared db_9b0fff_aiscompany). All access via SRN_ stored procedures, scoped to the hospital tenant.
3D & AR
Three.js (ESM importmap three@0.160.0) with STLLoader, OrbitControls and ARButton for WebXR hit-test placement.
DICOM
Client-side dicomParser (pure JS via CDN) for non-PHI metadata and PHI-tag stripping.
Voice
Web Speech API — SpeechRecognition for hands-free questions and speechSynthesis to read answers back.
AI
SRN_ClaudeHelper routes to Groq or Claude/Anthropic, returns JSON, and scrubs PHI from free text as a safety net.
Key tables
| Table | Purpose |
|---|---|
| SRN_Hospitals | Tenant hospitals: name, city, contact, subscription tier and expiry. |
| SRN_Surgeons | Surgical team members linked to kit users, with specialty and registration number. |
| SRN_Patients | De-identified patient records: patient code, age, sex, plus hospital-internal display name and notes (never sent to AI). |
| SRN_DicomUploads | Uploaded DICOM/STL files with PHI-stripped metadata and the stored model path. |
| SRN_Procedures | Procedures with body region, description, STL model path, marker image path and template flag. |
| SRN_ProcedureSteps | Ordered safety-checklist steps per procedure: step number, title, instruction, safety note. |
| SRN_IntraOpSessions | Intra-op sessions tying a procedure and patient together, with start/end and voice-command log. |
| SRN_Subscriptions | Payment/subscription records; on paid status extends the hospital’s expiry by one year. |
SRN_usp_GetHospital, SRN_usp_GetDashboard,
SRN_usp_GetPatients, SRN_usp_GetProcedures,
SRN_usp_GetProcedureSteps, SRN_usp_StartIntraOpSession,
SRN_usp_LogVoiceCommand and SRN_usp_SaveSubscription.Who can do what
Hospital admin
Has the settings capability: manages surgeons, hospital settings and the annual subscription, and views the audit log.
Surgeon / staff
Prepares patients, sets up procedures, runs AR navigation, checklists and operative notes for their hospital.
Super admin
Shared-kit global/module super admin monitors the product centrally alongside the rest of the AR suite.
Simple annual plan
Rs.50,000 / hospital / year
One annual plan unlocks every module for the whole hospital and all its surgeons and residents. Each Razorpay payment keeps the plan live for one year from the payment date.
Billing & security
Payments are processed by Razorpay (PCI-DSS compliant). Orders carry notes.hospital_id so the HMAC-verified webhook attributes each payment to the correct hospital. Tax invoice available on request.
Future ideas (as on today)
- Volumetric DICOM segmentation — assisted, in-app segmentation of DICOM series into meshes, so hospitals no longer pre-segment STL files in separate tools.
- Improved AR registration — explore marker/image tracking and depth sensing for steadier overlay, always keeping the visualisation-aid (not certified-device) framing.
- Multi-language voice — hands-free anatomy answers and dictation in regional languages for wider theatre use.
- Structured note templates — per-specialty operative-note templates and export to the hospital record, keeping de-identification at the AI boundary.
- Training analytics — resident quiz history, weak-topic tracking and cohort dashboards for training programmes.
- Offline-friendly viewer — PWA caching of models and checklists for use in theatres with limited connectivity.