FACES: A Deep-Learning-Based Parametric Model to Improve Rosacea Diagnoses

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Rosacea is a chronic inflammatory skin condition characterized by facial redness and visible blood vessels.
• Current diagnostic methods rely on visual assessment, which is subjective and leads to significant inter-clinician variability.
• Advancements in artificial intelligence (AI) show promise for objective and consistent disease detection in medical imaging.

Purpose of the Study:

• To develop and evaluate an AI-based system for the efficient and accurate identification and classification of rosacea.
• To compare the performance of the novel AI system against individual machine learning models and traditional diagnostic methods.

Main Methods:

• Developed a 'five accurate CNNs-based evaluation system' (FACES) utilizing five top-performing convolutional neural network (CNN) models.
• Trained and validated 19 CNN models on image datasets for rosacea detection.
• Selected the five best models based on accuracy to form the FACES system, incorporating a majority rule for classification.

Main Results:

• The FACES system demonstrated superior performance compared to individual CNN models and the majority rule.
• FACES achieved the highest accuracy and sensitivity in rosacea detection.
• Specificity and precision of FACES were higher than most individual models, indicating robust diagnostic capability.

Conclusions:

• The FACES system offers a more accurate and consistent method for rosacea identification and classification.
• AI-driven approaches like FACES have the potential to enhance dermatological diagnostics.
• Future research should incorporate patient demographics and clinical comparisons to further refine the system.