Automated triage of cancer-suspicious skin lesions with 3D total-body photography
AI-Powered 3D Total Body Photography for Early Skin Cancer Detection
This research details the 'ISIC 2024 – Skin Cancer Detection with 3D-TBP' grand challenge, showcasing the development of machine learning models for automated triage of cancer-suspicious skin lesions using 3D total body photography. The winning model significantly outperforms previous approaches by incorporating intra-patient context and various metadata, achieving high diagnostic accuracy and substantially reducing the number of lesions requiring expert evaluation. This advancement paves the way for more efficient and accessible skin cancer surveillance.
Executive Impact: Key Performance Indicators
Our analysis highlights the quantitative advantages of integrating AI into dermatological workflows.
AUROC (Area Under the Receiver Operating Characteristic)
NNT (Number Needed to Triage) for 80% Sensitivity
Reduction in lesions needing expert review
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
The winning AI model leverages intra-patient context, mirroring clinical practice, to identify outlier lesions. This significantly improves diagnostic accuracy (AUC = 0.967 vs. 0.956 without context) and reduces the NNT for 80% sensitivity from 72.68 to 50.57. This highlights the importance of considering a lesion in relation to all other lesions on a patient, rather than in isolation.
The study demonstrates the superior performance of multimodal AI models, combining image data (tiles), WB360 'appearance' metadata, and basic 'demographics' metadata. Image-only models underperformed, indicating that black-box feature extraction from images alone is suboptimal. The integration of diverse data types is crucial for robust skin cancer detection.
The AI system can triage 80% of skin cancers by reviewing only 1 in 51 lesions, or 90% by reviewing 1 in 98. This represents a 95% reduction in lesions requiring expert assessment compared to previous methods, offering substantial improvements in clinical workflow efficiency for dermatology specialists and potentially increasing access to care.
95%
Reduction in lesions needing expert review for 80% sensitivity
AI Model Workflow for 3D TBP Analysis
3D Total Body Photography Acquisition
→
Lesion Tile Extraction & Metadata Gathering
→
Image-based Feature Extraction (Neural Networks)
→
Metadata-based Feature Engineering (Patient Context)
→
Gradient Boosting Model Training
→
Automated Risk Score Generation
→
Triage for Expert Dermatological Review
| Feature Set | PAUC>80% TPR | AUC | NNT 80% SE | Key Advantages |
|---|---|---|---|---|
| Full Model (Tiles + All Metadata) | 0.173 | 0.967 | 50.57 |
|
| Tiles + Basic + WB360 Metadata | 0.165 | 0.956 | 72.68 |
|
| Tiles Only | 0.142 | 0.922 | 143.21 |
|
| Marchetti et al. (Benchmark) | 0.032 | 0.704 | 874.27 |
|
Real-World Impact: Streamlining Melanoma Surveillance
Challenge: A high-risk patient with hundreds of nevi requires frequent full-body skin examinations. Manually identifying and tracking every suspicious lesion is time-consuming and prone to human error, leading to potential delays in diagnosis and increased dermatologist workload.
Solution: Implementing the ISIC'24 winning AI model, integrated with 3D total body photography, automates the initial triage. The system analyzes all visible lesions, identifies those with a high-risk score, and flags them for dermatologist review, leveraging intra-patient context to prioritize truly atypical lesions.
Results: The dermatologist's review time is significantly reduced, focusing only on the ~5% of lesions flagged by the AI. This leads to earlier detection of new or changing melanomas, improved patient outcomes, and optimized clinic efficiency, allowing dermatologists to see more patients or dedicate more time to complex cases.
Advanced ROI Calculator
Estimate the potential return on investment for integrating AI-powered dermatological screening into your practice or healthcare system.
Potential Annual Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A phased approach to integrate AI-powered 3D TBP triage into your clinical workflow.
Phase 1: Pilot Program & Data Integration (3-6 Months)
Establish a pilot program with a small group of clinicians. Integrate 3D TBP system with existing EHR. Secure data pipelines and ensure compliance. Initial training for staff on new workflows.
Phase 2: Model Customization & Validation (6-12 Months)
Refine AI model parameters based on local patient demographics and clinical protocols. Conduct internal validation studies to confirm diagnostic performance and NNT metrics. Gather clinician feedback for workflow optimization.
Phase 3: Scaled Deployment & Ongoing Monitoring (12+ Months)
Expand deployment across relevant departments/clinics. Implement continuous monitoring of AI performance and patient outcomes. Establish a feedback loop for model retraining and updates to ensure sustained accuracy and efficiency.
Ready to Transform Your Practice with AI?
Book a personalized session with our AI specialists to explore how these insights can be tailored to your enterprise needs.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat