AI Model Trails Expert Skin Lesion Readers

Overview

A study published in JAMA Dermatology found that an AI model outperformed physicians with less than 3 years of dermoscopy experience in diagnosing skin lesions but did not surpass dermatologists with over 10 years of experience. The AI showed superior performance in benign vs malignant discrimination but lagged in multiclass diagnostic accuracy.

Background

Skin lesions can indicate a range of dermatological disorders, including skin cancer, making accurate diagnosis crucial for effective treatment. The integration of artificial intelligence in dermatology has the potential to enhance diagnostic accuracy, particularly among less experienced practitioners.

Data Highlights

Key Findings

• The unimodal PanDerm model achieved 72% accuracy, outperforming physicians with less than 3 years of experience.
• Dermatologists with over 10 years of experience had the highest diagnostic accuracy at 74%.
• The unimodal model had the highest balanced accuracy in binary discrimination at 0.82.
• Physicians' overall sensitivity and specificity were 66% and 65%, respectively.
• The multimodal model performed worse than the unimodal model despite having additional clinical information.
• AI tools may serve as decision-support systems for less experienced clinicians.

Clinical Implications

The findings indicate that AI models can assist in diagnosing skin lesions but do not replace the expertise of seasoned dermatologists.

Conclusion

The study highlights the performance of AI in dermatology while emphasizing the continued importance of human expertise.

Related Resources & Content

1. JAMA Dermatology, 2023 -- AI Model Trails Expert Skin Lesion Readers
2. conexiant, 2025 -- AI Model Differentiates BCC vs cSCC Subtypes
3. DIGITAL HEALTH, 2026 -- Enhancing skin lesion classification using a Tri-Path Attention Stacked Ensemble architecture with Cohen’s Kappa Proportioned Averaging
4. the asco post, 2025 -- Pathology Machine-Learning Models and Diagnosis of Nonmelanoma Skin Cancers in Resource-Limited Settings
5. Cutaneous melanoma: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up - PMC
6. Frontiers in Immunology — A two-stage workflow for vitiligo diagnosis: clinical characteristic classification and large language model (LLM)–based report generation
7. Cutaneous melanoma: ESMO Clinical Practice Guideline for diagnosis, treatment and follow-up - PMC
8. Limits of Artificial Intelligence Models for Skin Cancer Diagnosis in Realistic Settings
9. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension | Nature Medicine