UC59: Dermatology Triage

The Problem

Dermatology has severe access barriers: wait times exceed 30 days in many regions; benign skin lesions clog referral pipelines, delaying evaluation of potentially malignant lesions. AI-powered preliminary classification of patient-submitted lesion photos can triage urgency and reduce unnecessary referrals, improving access for high-risk lesions.

What the Agent Does

Patient or PCP submits photo of skin lesion via patient portal or secure messaging
Agent performs preliminary classification using computer vision: analyzes lesion appearance (size, border irregularity, color variation, surface characteristics)
Applies ABCDE criteria (Asymmetry, Border irregularity, Color variation, Diameter >6mm, Elevation/Evolution) and machine learning model trained on melanoma vs. benign lesions
Generates triage category: "Routine referral (likely benign mole)" vs. "Urgent referral (suspicious for melanoma; recommend evaluation within 2 weeks)"
DOES NOT diagnose; explicitly disclaims "This is a preliminary classification only; a dermatologist must evaluate"
Provides patient with guidance: if urgent, directs to fastest-available dermatology appointment or urgent care
Routes referral to dermatology with triage flag; urgent cases prioritized
Dermatologist reviews photo + triage classification and makes final determination

Data Requirements

Primary Data Sources: Patient-submitted lesion photos, patient demographics (age, skin type, family history of melanoma), ABCDE assessment criteria, computer vision model trained on melanoma vs. benign lesions
Classification: PHI/PII (patient identifiers, lesion photos)
Quality Metrics: Sensitivity for melanoma >90% (capture true high-risk lesions); specificity >70% (avoid false positives); triage accuracy validated against dermatologist assessment >85%
Complexity: Very High (medical image deep learning, melanoma classification, regulatory scrutiny)
Systems: Epic patient portal, DermEngine (dermatology AI platform)

Score Breakdown

Criterion	Weight	Score (1-5)	Weighted
Time Recaptured	15%	4	0.60
Error Reduction	10%	3	0.30
Cost Avoidance	10%	3	0.30
Strategic Leverage	5%	3	0.15
Data Availability	15%	3	0.45
Process Clarity	15%	3	0.45
Ease of Implementation	10%	3	0.30
Fallback Available	10%	3	0.30
Audience (Int/Ext)	10%	3	0.30
Composite	100%		3.25

Why It Scores Well

Access Improvement: Reduces dermatology referral burden by 30 to 40%; urgent cases prioritized
Patient Safety: Melanoma detection improves; early intervention improves outcomes
Regulatory Fit: Triage is classification, not diagnosis; FDA likely does not require premarket approval for triage-only tools; if tool recommends treatment, FDA review required
Scalability: Applies to all patients with suspicious skin lesions
Evidence Base: AI melanoma detection algorithms have achieved >90% sensitivity in published studies

Regulatory Alignment

FDA: Dermatology AI tools that provide "triage" (classify risk) may be exempt from premarket approval; tools that "diagnose" (recommend biopsy or treatment) typically require 510(k) review. Intended use statement critical.
State Medical Practice: Tool must not perform independent diagnosis; must be classified as clinical decision support
State Telemedicine Laws: Photo submission may trigger telemedicine licensing requirements depending on state and follow-up care plans
HIPAA: Lesion photos are PHI; must be encrypted and securely stored

Sprint Factory Fit

4 to 5 sprints (computer vision model training/validation, portal integration, triage workflow, FDA premarket pathway, clinical governance review)

Sprints Required: 4 to 5 sprints (computer vision model training/validation, portal integration, triage workflow, FDA premarket pathway, clinical governance review)
Timeline: 10 to 12 weeks from requirements to pilot; potentially 6 to 12 months if FDA 510(k) required
Governance: Dermatology medical director validation of triage accuracy in first 100 lesions; validate sensitivity >90% and specificity >70%
Success Metrics: Triage accuracy >85%; sensitivity >90%; dermatology wait times reduced >20%; urgent lesion detection rate >85%; patient satisfaction >4.0/5

Comparable Implementations

DermEngine: AI dermatology decision support
Research: Melanoma detection algorithms (>90% sensitivity published)
FDA 510(k) database: Skin lesion classification tools

Deploy This Use Case with the Sprint Factory

From zero to a governed, production agent in 6 weeks.

Sprint Factory Schedule a Briefing

Related Use Cases

On-Demand Clinical

AI Clinical Scribe & Note Generator

Score: 4.3/5.0

On-Demand Clinical

Medical Knowledge Base Search

Score: 3.75/5.0

On-Demand Clinical

Patient History Summary ("Highlight Reel")

Score: 3.6/5.0