WEDNESDAY, May 14, 2025 (HealthDay News) -- A robust, deep learning model has been developed for the classification of melasma severity, according to a study published online April 16 in Clinical, Cosmetic and Investigational Dermatology.Jun Zhang, from Wuhan First Hospital in China, and colleagues developed an artificial intelligence-assisted real-time melasma severity classification framework based on deep learning and 1,368 anonymized facial images, obtained from clinically diagnosed patients. Six convolutional neural network architectures were trained and evaluated after image preprocessing and Melasma Area and Severity Index-based labeling, and model performance was assessed.The researchers found that the best performance was achieved by GoogLeNet, with accuracy of 0.755 and an F1-score of 0.756. Across severity levels, area under the curve values reached 0.93, 0.86, and 0.94 for mild, moderate, and severe, respectively. GoogLeNet's superior feature attribution was confirmed in a Layer-wise Relevance Propagation analysis."This study demonstrates that deep learning provides a transformative solution for improving the accuracy, consistency, and clinical applicability of melasma severity assessment," the authors write.Abstract/Full Text.Sign up for our weekly HealthDay newsletter
WEDNESDAY, May 14, 2025 (HealthDay News) -- A robust, deep learning model has been developed for the classification of melasma severity, according to a study published online April 16 in Clinical, Cosmetic and Investigational Dermatology.Jun Zhang, from Wuhan First Hospital in China, and colleagues developed an artificial intelligence-assisted real-time melasma severity classification framework based on deep learning and 1,368 anonymized facial images, obtained from clinically diagnosed patients. Six convolutional neural network architectures were trained and evaluated after image preprocessing and Melasma Area and Severity Index-based labeling, and model performance was assessed.The researchers found that the best performance was achieved by GoogLeNet, with accuracy of 0.755 and an F1-score of 0.756. Across severity levels, area under the curve values reached 0.93, 0.86, and 0.94 for mild, moderate, and severe, respectively. GoogLeNet's superior feature attribution was confirmed in a Layer-wise Relevance Propagation analysis."This study demonstrates that deep learning provides a transformative solution for improving the accuracy, consistency, and clinical applicability of melasma severity assessment," the authors write.Abstract/Full Text.Sign up for our weekly HealthDay newsletter