The AI microservice for differential diagnosis reasoning
DDXaid plugs into any clinical app to turn structured findings into an explainable, ranked differential list with rule-in/out criteria and concise rationale.
Why teams choose DDXaid
Explainable DDx, workflow-ready, and API-first
Built for clinicians, platforms, and training
Use DDXaid as a reasoning layer that enhances care and education.
Product Overview
DDXaid is an AI-powered microservice that turns structured or semi-structured clinical findings (e.g., ultrasound reports, labs, vitals) into an explainable, ranked differential diagnosis list with rule-in/out criteria, guideline references, and concise reasoning summaries.
Vision
Become the industry-standard DDx reasoning layer — embeddable in any clinical app, EMR, or AI assistant to suggest and prioritize diagnoses, support structured rule-out workflows, and provide guideline-aligned explanations for trust, auditability, and training.
Product Objectives
How it works
1) Analyze
/ddx/analyze
2) Update
/ddx/{session_id}/update
3) Continue
/ddx/{session_id}
Core Features
DDx Reasoning Engine
Hybrid approach for trustworthy results
- Rules-based library for deterministic baseline.
- Bayesian-weighted probability scoring.
- LLM-bounded rationale summarization (concise and explainable).
Rule-Out Workflow
Checkboxable criteria with instant feedback
- Each candidate includes rule-in/out criteria as checkboxes.
- Clinicians update criteria; statuses and scores recalc instantly.
- Concise explanations like “Ruled out due to normal Doppler.”
Explainability Layer
Structured outputs built for trust
- likelihood_score (0–1 normalized)
- rationale_summary (concise reasoning)
- guideline_refs (e.g., “SMFM #52, Diagnosis section”)
- suggested_tests (optional guidance)
API Architecture
Simple endpoints for fast integration
/ddx/analyze— generate DDx from findings./ddx/{session_id}/update— apply clinician decisions./ddx/{session_id}— fetch current state for audit/continuation.
Audit & Compliance
Traceable interactions, ready for review
- Record timestamp, actor, and action type for each change.
- Enable traceability for HIPAA/SOC2 compliance expectations.
Approach to AI Usage
We use a hybrid approach: a curated, deterministic rule library for ranking and an AI model only for concise rationale text. This keeps results safe, consistent, and explainable while still helping uncover missed considerations.
Rule-first ranking
Deterministic, auditable, guideline-aligned
- Curated rule-in/out criteria and weights per condition.
- Transparent scoring you can tune and review.
- Findings normalization increases recall across synonyms.
AI for rationale only
Summaries, not decision-making
- Generates concise reasoning and key points for documentation.
- Does not change ranking or scores.
- References and suggested tests remain grounded in rules.
Uncovering missed diagnoses
Surface candidates a clinician may not consider
- When input findings match a condition’s rules, it is suggested even if not explicitly ordered.
- Canon mapping recognizes varied terms (e.g., “Absent CSP” synonyms) to avoid gaps.
- Optional “Other considerations (AI)” lane can propose items outside the rule library, clearly labeled and not ranked.
Coverage & evolution
Continuous expansion of the library
- We add specialties and conditions over time.
- Weights calibrated with clinical feedback and literature.
- Future: rules from YAML/DB to enable rapid updates.
Clinical decision support; not a diagnosis.
Example: MFM Ultrasound Consult
Enter findings like “EFW <10th percentile,” “Elevated UA PI,” “AFI normal.” DDXaid returns candidates such as Fetal Growth Restriction and Preeclampsia Spectrumwith structured criteria. As you check off rule-in/out boxes, the likelihoods and rationale update instantly for inclusion in the consult note.
Docs and SDKs
Get started with our API in minutes. SDKs and examples coming soon.