Glaucoma Agentic RAG & Decision Framework

Patient Clinical Ingestion

Measured IOP (Tonometry) 18.0 mmHg

10 mmHg (Low) 21 mmHg (Limit) 45 mmHg (High)

Vertical Cup-to-Disc Ratio (CDR) 0.45

0.1 (Normal) 0.5 (Suspect) 0.95 (Severe)

Central Corneal Thickness (CCT) 545 µm

400 µm (Thin) 545 µm (Average) 680 µm (Thick)

Patient Age 58 Years

18 Yrs 60 Yrs (Risk Threshold) 95 Yrs

First-Degree Glaucoma Family History

Co-morbid Systemic & Vascular Risk Factors

Chronic Migraines Raynaud's (Vasospasm) Hypertension Diabetes Mellitus

Retinal Optic Disc Image Input

Normal Disc (CDR 0.3)

Glaucomatous Disc (CDR 0.8)

Progression Hazard Risk

Corneal-Corrected IOP

- mmHg

Correction: +0.0 mmHg

Predicted Target IOP

clinical recommendation

Initialize patient inputs and click "Execute Agent Reasoning" to begin.

No guidelines retrieved. Run analysis to trigger the Retrieval Agent.

Clinical summary will be synthesized here.

Multi-Agent Decision Pipeline (Mermaid)

graph TD
    Input[Patient Clinical Metrics / Image] --> Triage[Triage Agent]
    Triage -->|Extracts Keywords and Flags| RAG_Query[Retrieval Agent]
    RAG_Query -->|TF-IDF Guideline Scan| LocalDB[(Guidelines Database)]
    LocalDB -->|Top-N Matches and Scores| RAG_Query
    RAG_Query -->|Citations and Evidence Chunks| Reasoning[Clinical Reasoning Agent]
    Triage -->|Risk Factors and Raw IOP| Reasoning
    Reasoning -->|CCT Correction Formula and CoT Evaluation| Consensus[Consensus and Report Agent]
    Consensus -->|Generate Target IOP and Risk Index| Report[Diagnostic Report and Visual Dashboard]

Live Rendered Pipeline:

Framework Formulation (LaTeX)

% --- IOP Central Corneal Thickness Correction Algorithm ---
\text{IOP}_{\text{corrected}} = \text{IOP}_{\text{measured}} + \frac{545 - \text{CCT}}{10} \times 0.6 \, \text{mmHg}

% --- Multi-Agent Clinical Risk Scoring Metric ---
R_{\text{glaucoma}} = w_p \cdot f_p(\text{IOP}_{\text{corrected}}) + w_s \cdot \text{CDR}_{\text{index}} + w_c \cdot I(\text{CCT} < 520) + \sum w_a \cdot A_i
\quad \text{where} \; R \in [0, 15]

% --- Agent Target Pressure Decision Boundary ---
\text{Target IOP} = \begin{cases} 
      \min(14, \text{IOP}_{\text{corrected}} \times 0.70) & R \ge 8 \; (\text{High Risk}) \\
      \min(18, \text{IOP}_{\text{corrected}} \times 0.78) & 5 \le R < 8 \; (\text{Moderate Risk}) \\
      20 & R < 5 \; (\text{Mild/Low Risk})
   \end{cases}

Chapter Snippet Idea:

"Figure 4.2 illustrates how a decoupled multi-agent decision support loop avoids single-prompt failures. The Triage Agent extracts physiological parameters, sending structured markers to the Retrieval Agent. The Retrieval Agent executes RAG, extracting targeted medical consensus. Finally, the Clinical Reasoning Agent uses Chain-of-Thought mathematical bounds to calculate corneal-adjusted intraocular pressure and synthesizes risks into a target pressure recommendation."