All published articles of this journal are available on ScienceDirect.
Artificial Intelligence in Dermatology: Assessing Predictability in Clinical Diagnosis
Ahmed Abdel Khabir2
Doaa Shokry Al Emam3
Sara Hamdy Fouad2
Authors Info & Affiliations
Abstract
Introduction
The use of Artificial Intelligence (AI) for image-based diagnosis in dermatology is increasing rapidly. The clinical accuracy of AI in diagnosing different skin conditions remains under evaluation. This study aimed to evaluate the diagnostic performance of an AI application in comparison to confirmed clinical diagnoses by dermatologists.
Method
A cross-sectional study was carried out on 400 patients with different skin conditions, including acne, alopecia, eczema, pigmentary disorders, psoriasis, immunological disorders, tumors, infections, and infestations. The study analyzed AI-based predictions using the Tibot AI application, comparing them against dermatologists’ diagnoses.
Results
The AI application demonstrated high diagnostic accuracy for certain dermatological conditions such as adnexal disorders (AUC 0.93–0.98), pigmentary disorders (AUC 0.88–0.94), and cutaneous tumors (AUC 0.87–0.95). Sensitivity for adnexal disorders was 88.9% (top one) and 94.4% (top three), and for Pigmentary disorders, it was 75.8% and 87.9% for top one and top-three predictions, respectively.
However, AI performance was lower for immunological disorders (31.3% sensitivity) and cutaneous infestations (22.2%). Overall accuracy improved across all conditions when considering the top-three predictions.
Discussion
Tibot AI-application demonstrated high diagnostic accuracy for conditions with distinct morphological features such as adnexal, pigmentary disorders, and cutaneous tumors. It showed lower sensitivity for immunological disorders and infestations, indicating the need for further AI training with more diverse datasets.
Conclusion
AI-based diagnostic accuracy improved significantly when considering the top-three diagnoses, indicating its value as a differential diagnostic tool. It showed promising accuracy in adnexal, pigmentary disorders, and cutaneous tumors. However, it is less robust for immunological skin diseases and infections, highlighting the need for further refinement.