Artificial Intelligence and Its Effect on Dermatologists' Accuracy in Dermoscopic Melanoma Image Classification:
Roman C Maron1, Jochen S Utikal2,3, Achim Hekler1
1Digital Biomarkers for Oncology Group (DBO), National Center for Tumor Diseases (NCT), German Cancer Research Center (DKFZ), Heidelberg, Germany.
Journal of Medical Internet Research
|September 11, 2020
Summary
Artificial intelligence (AI) significantly improved dermatologists' accuracy in distinguishing melanoma from nevi. AI support enhanced diagnostic performance, aiding early detection of skin cancer.
Area of Science:
- Dermatology
- Medical Imaging
- Artificial Intelligence
Background:
- Early melanoma detection is crucial but challenging.
- Artificial intelligence (AI) shows promise in classifying skin lesions.
- AI algorithms may enhance dermatologist diagnostic capabilities.
Purpose of the Study:
- To evaluate if AI support improves dermatologist accuracy in differentiating melanoma from nevi.
- To assess AI's impact on overall diagnostic performance in image-based discrimination.
Main Methods:
- Twelve dermatologists classified 1200 dermoscopic images, first independently, then with AI (CNN) support.
- Dermatologists rated their confidence in each diagnosis.
- Performance metrics (sensitivity, specificity, accuracy) were compared between conditions.
Main Results:
- AI support significantly increased dermatologist sensitivity (59.4% to 74.6%) and accuracy (65.0% to 73.6%).
- Specificity remained largely unchanged (70.6% to 72.4%).
- Dermatologists corrected more AI errors than AI corrected dermatologist errors, and confidence levels adjusted with AI feedback.
Conclusions:
- AI support enhances dermatologist accuracy in melanoma-nevus discrimination.
- AI-based tools can aid clinicians in skin lesion classification.
- Further real-life studies integrating AI with clinical data are warranted.
