12 Aug 2026 · concept doc
CLD + LT platform architecture
Twenty-module journey from intake to remote follow-up. Source: Complete AI CLD Decision Trees (local Word) — a development framework, not a deployable protocol.
Exact diagnostic/treatment thresholds must be version-controlled against current institutional / national / international guidelines before any clinical use. Deterministic rules stay deterministic; ML only where it adds validated value.
How the platform should think
- Guideline rules that are safer as rules → keep as rules.
- Predictive AI only where discrimination / calibration / efficiency is validated.
- Every high-stakes output → clinician checkpoint. No autonomous prescribing or intraoperative action.
Master journey (20 modules)
| # | Module | Job |
|---|---|---|
| 1 | Patient Intake and Triage | Safe entry; red flags before routine CLD work |
| 2 | Intelligent Liver History | Structured history + etiologic risk profile |
| 3 | Physical Examination | Standard phenotype / decompensation signs |
| 4 | Diagnostic Investigations | Tiered labs/imaging — not indiscriminate testing |
| 5 | CLD Diagnostic Engine | Etiology + stage + complications (separately) |
| 6 | Cirrhosis / Portal Hypertension | Compensated vs decompensated routing |
| 7 | Etiology-Specific Treatment | Disease pathway for clinician review |
| 8 | Cirrhosis Complication Mgmt | Bleed / HE / SBP / HRS / ACLF submodules |
| 9 | HCC Surveillance & Treatment | Surveillance → MDT options |
| 10 | Prognostic Engine | Scores + longitudinal risk |
| 11 | LT Referral Engine | When to refer (not candidacy) |
| 12 | Liver Transplant Evaluation | MDT completeness / list / defer / decline |
| 13 | Surgical & Anatomical Planning | Technical complexity for surgeon review |
| 14 | Donor and Graft Selection | DDLT/LDLT match; GRWR / donor safety |
| 15 | Preoperative Optimization | Modifiable risk + readiness |
| 16 | Intraoperative Intelligence | Alerts only — no autonomous action |
| 17 | Early Post-LT Monitoring | POD trajectories (EAD/PNF/vascular…) |
| 18 | Post-LT Complication Detection | Branch diagnostics when deteriorating |
| 19 | Long-Term Follow-up | Graft / IS / metabolic / QoL |
| 20 | Patient-Facing AI / Remote | Guarded triage + education; escalate to clinician |
Suggested MVP from the same doc
CLD Navigator first — not the full 20:
- Diagnosis / etiology
- Fibrosis / cirrhosis stage
- Complication detection
- HCC surveillance
- Prognosis
- LT referral trigger
Clinician dashboard: current state + missing data + risks + next module + reason.
How this maps to what we already built
| Already shipping / local | Maps to |
|---|---|
| Screening / sheet intake (Matching) | Modules 1–2, parts of 12/14 |
| Hard GRWR / remnant / age match | Module 14 (LDLT safety rules) |
| Recipient + donor checklists (Protocols) | Modules 12–15 |
| Rounds clinical capture (separate product) | Modules 2–3, 17–19 documentation layer |
| Fellow field manual (local PDF) | Human protocol spine under Modules 15–18 |
Build order stay simple: finish sheet + hard match + Selva answers → then CLD Navigator MVP modules → only later donor/graft AI and intraoperative ideas.