Diagnosis Support

Overview

MedMitra provides AI-powered diagnostic support by analyzing clinical data and generating evidence-based diagnosis recommendations. The system considers patient history, lab results, imaging findings, and clinical presentation to suggest primary diagnoses with supporting evidence.

Diagnostic suggestions are for clinical decision support only and must be validated by qualified healthcare professionals. Not intended for direct patient diagnosis without physician review.

Diagnosis Model

Diagnoses are structured with comprehensive supporting data:

class Diagnosis(BaseModel):
    primary_diagnosis: str
    icd_code: Optional[str] = None
    description: str
    confidence_score: float = Field(ge=0.0, le=1.0)
    supporting_evidence: List[str]

Source: backend/models/data_models.py:70-75

Model Components

Primary Diagnosis

The most likely diagnosis based on available evidence

ICD Code

International Classification of Diseases code for billing and records

Description

Detailed explanation of the diagnosis and clinical presentation

Supporting Evidence

Specific findings that support the diagnosis

Generation Process

Diagnoses are generated from SOAP notes:

SOAP Note Input

Complete SOAP note provides clinical context

Evidence Analysis

LLM analyzes objective findings and assessment

Diagnosis Formulation

Primary diagnosis determined with supporting evidence

ICD Code Mapping

Appropriate ICD-10 code assigned

Confidence Scoring

System evaluates certainty of diagnosis

Implementation

async def _generate_diagnosis(self, state: MedicalAnalysisState) -> MedicalAnalysisState:
    """Generate primary diagnosis from SOAP note"""
    
    # Generate diagnosis using specialized prompt
    diagnosis_response = await self.llm_manager.generate_response(
        system_prompt=DIAGNOSIS_PROMPT,
        user_input="SOAP Note: " + state["soap_note"].model_dump_json()
    )
    
    # Create structured diagnosis
    diagnosis = Diagnosis(
        primary_diagnosis=diagnosis_response.get("diagnosis", ""),
        icd_code=diagnosis_response.get("icd_code"),
        description=diagnosis_response.get("description", ""),
        confidence_score=diagnosis_response.get("confidence_score", 0.8),
        supporting_evidence=diagnosis_response.get("supporting_evidence", [])
    )
    
    state["primary_diagnosis"] = diagnosis
    state["processing_stage"] = "diagnosis_generated"
    return state

Source: backend/agents/medical_ai_agent.py:186-207

Diagnostic Reasoning

The AI considers multiple factors:

Clinical Presentation

Symptoms: Pattern and duration of symptoms
Patient Demographics: Age, gender, risk factors
Medical History: Previous conditions and treatments
Temporal Progression: Acute vs chronic presentation

Objective Evidence

Laboratory Findings

Abnormal values indicating specific conditions
Patterns across multiple tests
Severity markers (inflammatory markers, cell counts)
Diagnostic biomarkers for specific diseases

Imaging Results

Radiographic patterns characteristic of diseases
Anatomical abnormalities
Comparison with previous studies
Severity and extent of findings

Clinical Context

Provider’s initial assessment
Physical examination findings
Response to previous treatments
Epidemiological factors

Example Diagnosis

Here’s a complete diagnosis output:

{
  "primary_diagnosis": "Community-Acquired Pneumonia",
  "icd_code": "J18.9",
  "description": "Acute bacterial or atypical pneumonia acquired outside of healthcare settings. Patient presents with respiratory symptoms, elevated inflammatory markers, and radiographic evidence of pulmonary infiltrates. Clinical presentation and bilateral interstitial pattern suggest atypical pneumonia, possibly mycoplasma or viral etiology.",
  "confidence_score": 0.87,
  "supporting_evidence": [
    "Bilateral interstitial infiltrates on chest X-ray",
    "Elevated WBC count (12,500/μL) indicating infection",
    "Elevated C-reactive protein (45 mg/L) indicating inflammation",
    "Clinical symptoms of cough and dyspnea",
    "No evidence of pleural effusion or consolidation typical of bacterial pneumonia",
    "Patient age and demographics consistent with atypical pneumonia"
  ]
}

Confidence Scoring

The confidence score reflects diagnostic certainty:

High Confidence (0.9 - 1.0)

Multiple supporting findings align clearly
Classic presentation of well-defined condition
Pathognomonic findings present
No contradictory evidence

Moderate Confidence (0.7 - 0.9)

Several supporting findings present
Typical but not definitive presentation
Some findings support diagnosis
Minor inconsistencies possible

Low Confidence (Below 0.7)

Limited supporting evidence
Atypical presentation
Multiple differential diagnoses possible
Requires additional investigation

Supporting Evidence

Each diagnosis includes specific evidence:

supporting_evidence = [
    "Lab finding: Elevated WBC indicating infection",
    "Imaging: Bilateral infiltrates on CXR",
    "Clinical: Progressive respiratory symptoms",
    "History: Recent exposure in healthcare setting",
    "Physical: Decreased O2 saturation"
]

Evidence is categorized by:

Laboratory data: Specific abnormal values
Imaging findings: Radiographic patterns
Clinical presentation: Symptoms and signs
Patient history: Risk factors and exposures

ICD-10 Coding

The system provides appropriate ICD-10 codes:

Code Structure

Category: Disease classification (e.g., J00-J99 for respiratory)
Specificity: Detailed code for precise diagnosis
Laterality: Left/right/bilateral when applicable
Episode: Initial encounter, subsequent, sequela

Common Codes Generated

Respiratory Infections

J18.9 - Pneumonia, unspecified J20.9 - Acute bronchitis

Cardiovascular

I10 - Essential hypertension I25.10 - Atherosclerotic heart disease

Metabolic

E11.9 - Type 2 diabetes mellitus E78.5 - Hyperlipidemia

Musculoskeletal

M25.50 - Joint pain, unspecified M79.3 - Panniculitis

Integration with Medical Insights

Diagnosis is part of the complete insights package:

class MedicalInsights(BaseModel):
    case_summary: CaseSummary
    soap_note: SOAPNote
    primary_diagnosis: Diagnosis  # Diagnosis component
    overall_confidence_score: float
    generated_at: datetime

Source: backend/models/data_models.py:96-104

Differential Diagnosis (Future Feature)

The system is designed to support differential diagnoses:

class DifferentialDiagnosis(BaseModel):
    condition: str
    probability: float = Field(ge=0.0, le=1.0)
    reasoning: str
    distinguishing_factors: List[str]

Source: backend/models/data_models.py:77-81

Differential diagnosis generation is currently commented out in the codebase but designed for future implementation.

Planned Differential Features

Ranked Alternatives: Top 3-5 alternative diagnoses
Probability Scores: Likelihood of each diagnosis
Distinguishing Factors: Key differences between diagnoses
Recommended Tests: Investigations to differentiate

Investigation Recommendations (Future Feature)

Planned feature for diagnostic workup:

class InvestigationRecommendation(BaseModel):
    investigation_type: str
    urgency: Literal["urgent", "routine", "follow-up"]
    rationale: str
    expected_findings: Optional[str] = None

Source: backend/models/data_models.py:83-87

Recommendation Categories

Urgent: Immediate testing required for patient safety
Routine: Standard workup for diagnosis confirmation
Follow-up: Monitoring tests after treatment initiation

Treatment Recommendations (Future Feature)

Planned therapeutic guidance:

class TreatmentRecommendation(BaseModel):
    treatment_type: str
    description: str
    dosage: Optional[str] = None
    duration: Optional[str] = None
    precautions: Optional[List[str]] = None

Source: backend/models/data_models.py:89-94

Accessing Diagnosis Data

Retrieve diagnosis through the case API:

GET /cases/cases/{case_id}

Response:
{
  "case": {...},
  "ai_insights": {
    "primary_diagnosis": {
      "primary_diagnosis": "Community-Acquired Pneumonia",
      "icd_code": "J18.9",
      "description": "Acute pneumonia...",
      "confidence_score": 0.87,
      "supporting_evidence": [
        "Bilateral infiltrates on CXR",
        "Elevated WBC count",
        "..."
      ]
    },
    "soap_note": {...},
    "case_summary": {...}
  }
}

Clinical Decision Support Workflow

Data Collection

Gather all clinical data (history, labs, imaging)

AI Analysis

System processes data and generates diagnosis

Physician Review

Doctor reviews AI suggestions and supporting evidence

Clinical Validation

Physician validates or modifies based on clinical judgment

Patient Care

Final diagnosis used for treatment planning

Limitations and Considerations

Important Limitations:

AI suggestions require physician validation
Cannot replace clinical examination and judgment
May miss rare or unusual presentations
Limited by quality and completeness of input data
Cannot account for all patient-specific factors
Not suitable for emergency/life-threatening situations without immediate physician oversight

Best Practices

For Optimal Diagnostic Support:

Complete Data Entry: Provide comprehensive case summaries
Quality Documents: Upload clear, complete lab and imaging reports
Clinical Context: Include relevant history and risk factors
Critical Review: Always validate AI suggestions clinically
Confidence Assessment: Pay attention to confidence scores
Supporting Evidence: Review all evidence listed
Patient Context: Consider individual patient circumstances

Use Cases

Primary Care

Quick reference for common conditions

Emergency Medicine

Rapid triage assistance

Specialist Referrals

Documentation for referral justification

Medical Education

Teaching tool for diagnostic reasoning

Quality Assurance

The system includes multiple validation layers:

Data Validation: Pydantic models ensure structure
Confidence Scoring: Flags uncertain diagnoses
Evidence Requirements: Diagnosis must have supporting evidence
ICD Code Validation: Codes checked against standard registry
Clinical Review: Physician validation required

Overview

Getting Started

Core Features

User Guide

AI Agents

​Overview

​Diagnosis Model

​Model Components

Primary Diagnosis

ICD Code

Description

Supporting Evidence

​Generation Process

​Implementation

​Diagnostic Reasoning

​Clinical Presentation

​Objective Evidence

​Example Diagnosis

​Confidence Scoring

​High Confidence (0.9 - 1.0)

​Moderate Confidence (0.7 - 0.9)

​Low Confidence (Below 0.7)

​Supporting Evidence

​ICD-10 Coding

​Code Structure

​Common Codes Generated

Respiratory Infections

Cardiovascular

Metabolic

Musculoskeletal

​Integration with Medical Insights

​Differential Diagnosis (Future Feature)

​Planned Differential Features

​Investigation Recommendations (Future Feature)

​Recommendation Categories

​Treatment Recommendations (Future Feature)

​Accessing Diagnosis Data

​Clinical Decision Support Workflow

​Limitations and Considerations

​Best Practices

​Use Cases

Primary Care

Emergency Medicine

Specialist Referrals

Medical Education

​Quality Assurance

​Next Steps

SOAP Notes

Case Management

Build docs developers (and LLMs) love

Overview

Diagnosis Model

Model Components

Generation Process

Implementation

Diagnostic Reasoning

Clinical Presentation

Objective Evidence

Example Diagnosis

Confidence Scoring

High Confidence (0.9 - 1.0)

Moderate Confidence (0.7 - 0.9)

Low Confidence (Below 0.7)

Supporting Evidence

ICD-10 Coding

Code Structure

Common Codes Generated

Integration with Medical Insights

Differential Diagnosis (Future Feature)

Planned Differential Features

Investigation Recommendations (Future Feature)

Recommendation Categories

Treatment Recommendations (Future Feature)

Accessing Diagnosis Data

Clinical Decision Support Workflow

Limitations and Considerations

Best Practices

Use Cases

Quality Assurance

Next Steps